Wireless body area networks (WBANs) have received a great deal of attention for facilitation of different applications including healthcare, sport activities, and personal assistance. These applications involve on-to-off body communication scenarios where physiological and activity data from a person are sent to the monitoring server. In such scenarios, the radio signal is affected by environmental factors as well as by human body (e.g., movements, structures, and tissues, etc.). As a result, typical wireless channel models do not fit well with WBAN environments.

For several years, many efforts have been made to identify the propagation characteristics in on-to-off body communication [1-4]. In indoor environments, the log-normal and Rayleigh distributions provide good fit to the on-to-off body channels for standing and for walking, respectively [1]. Smith et al. [2] showed that the on-to-off body channels for sleeping persons follow the Gamma distribution. The variation in the channel characteristic with the blockage of the line-of-sight path by the human body (i.e., shadowing effect) was investigated in [3]. A new statistical channel model for the path blockage was also introduced in [4]. Other investigations of the on-to-off body channel characteristics can be found in [5,6].

The channel characteristics, in fact, depend on various factors such as the environments, subjects (gender and body morphology), types, and positions of the antennas in addition to the body activities [7,8]. However, with the implicit assumption that the effects of the other factors are relatively negligible, the previous investigations focused on how the propagation channel characteristics change according to the body movements. Based on our preliminary experimental results, the on-to-off body channel characteristics vary considerably due to the transmitting antenna positions on the body while the other conditions are fixed. Efforts are ongo-ing to investigate the effects of antenna positions on the onbody channels [8], but to the best of our knowledge, no substantial propagation characterization and analysis of the on-to-off body channel has been conducted with respect to the antenna positions.

Recent experimental results (see [9] and references therein) showed that the power absorption is a dominant factor influencing the WBAN performance. Motivated by these results, we investigate the effect of body absorption losses (ALs) on the on-to-off body channel characteristic in this paper. The received signal strength (RSS) and the ALs are measured with respect to the on-body transmitting antenna positions through experiments and simulations. A statistical analysis is presented to characterize the on-to-off body channels as a function of the on-body transmitting antenna positions.

The rest of this paper is organized as follows. Section II describes the measurement setup and the power loss model. Section III analyzes the effect of the body AL on the on-tooff body channel with experimental and simulation results. Section IV concludes this paper.

The measurement system consisted of a transmitting/receiving (Tx/Rx) pair of planar inverted-F antennas (PIFAs) and a vector network analyzer (VNA; Rohde & Schwarz ZVB20, Germany). Fig. 1 shows the geometry of the PIFA. The PIFA is designed on a FR4 substrate with a relative permittivity of 4.4, and the total size of the substrate is 33×42 mm². The antenna has a dimension of 23×8×0.8 mm³. The PIFAs operated in the 2.45 GHz ISM band have an omnidirectional radiation pattern in free space (Fig. 2(a)) and have a directional pattern when mounted on the body (Fig. 2(b)).

A schematic diagram of the measurement setup is shown in Fig. 3. The Tx antenna was placed at one of three positions (ℓ₁, ℓ₂, ℓ₃) on the body, as shown in Fig. 3. ℓ₁ is the position of the heart, ℓ₂ is the position of the right upper-chest, and ℓ₃ is the position of the navel. The Rx antenna was placed 1 m away from the body, at the same height as the Tx antenna. Hence, the path loss was assumed to be constant regardless of the Tx antenna positions. The VNA generated a continuous wave signal of 0 dBm (1 mW) and sent a continuous wave signal through the Tx antenna at the 2.45 GHz ISM band. The VNA also operated in a single trace and single frequency mode to measure RSS in the timedomain. In a sweep time of 1 second, 1,601 samples were recorded for each Tx antenna position. The experiments were conducted on the body of a 1.76 m/73 kg male subject and a 1.60 m/50 kg female subject, standing motionless in an anechoic chamber. For the real experiments, the Tx antenna was attached to the skin, but a gap with a thickness less than 1 cm could occur due to the body morphology. The effects of multipath, fading, and shadowing from the human actions were neglected in the experiments. For each setting, 10 experiments were carried out. The SEMCAD-X simulations used the Duke model (1.77 m/72 kg) and Ella model (1.63 m/58 kg). The simulation settings were identical to those of the real experiments.

Consider a WBAN system in which a Tx antenna is mounted on ℓ_i and transmits a signal with a power of P_Tx. The RSS at the Rx antenna can then be expressed as

where d is the Tx-Rx distance separation, P_PL(d) is the path loss, P_a(d, ℓ_i) is the environmental loss by multipath, fading, and shadowing, and P_AL(ℓ_i) is the body AL when the Tx antenna is mounted on ℓ_i. In our measurement system, as the Tx-Rx distance is fixed to d = 1 m, and the human subjects are standing motionless in an anechoic chamber, the path loss P_PL(d) is considered constant and the environmental loss P_a(d, ℓ_i) is assumed to be negligibly small. Hence, (1) can be approximated as

where P₀ = P_Tx - P_PL(d), and it is assumed to be constant regardless of the Tx antenna positions. From (2), we intuitively know that the RSS is dependent on the body AL P_AL(ℓ_i).

The time-domain results of RSS measurements for the male and female subjects are shown in Fig. 4. The RSS can differ according to the Tx antenna positions. In the case of the male subject, RSS at ℓ₂ is higher than that at ℓ₃ by around 7 dB (Note that RSS at ℓ_i says that a value is measured at the Rx antenna when the Tx antenna is located at ℓ_i on the body). For the female subject, the largest difference between the RSS measurements with respect to the Tx antenna positions is approximately 3.5 dB. The average value of the RSS measurements is given with the simulation results and the AL in Table 1. The body AL P_AL(ℓ_i) for each Tx antenna position ℓ_i was measured with SEM-CAD-X simulations. According to Table 1, lower RSS is observed with greater AL, which means more power loss due to the body absorption. We note that the body structures of our subjects and the dielectric properties for tissues may not be same as those in the simulations, although the weights and heights of the subjects are similar to those of the Duke and Ella models.

Hence, the experiment and simulation results can differ. However, the order of RSS values in both experiments and simulations is identical. For the male subject, the lowest RSS is measured in both experiments and simulations when the Tx antenna is mounted on ℓ₃, where a total of 0.717 mW is absorbed. The RSS is found to be highest at ℓ₂, where the body absorption is lowest (0.568 mW). For the female subject, the highest and lowest ALs are observed at ℓ₁ and ℓ₂, respectively. As a result, the lowest RSS is measured when the Tx antenna is mounted on ℓ₁. However, the RSS at ℓ₃ is slightly higher than that at ℓ₂, although P_AL(ℓ₂) is lower than P_AL(ℓ₃). This is due to the curves on the female subject (body morphology). We will discuss on this in the following subsection.

The total AL by the human body depends on temperature, frequency, and the dielectric properties of tissues around the Tx antenna [10,11]. In our investigation, the change in the dielectric properties according to the Tx antenna positions only makes a difference because the effects of temperature and frequency remain constant. Although the temperatures can differ at various body (Tx antenna) positions, note that the AL due to the body temperature is considered constant regardless of the positions since the temperature differences are insignificant. The dielectric properties are defined in terms of relative permittivity є_r, and conductivity σ, and their numerical values for the simulations are described in Table 2. The absorption is well-known to depend more on conductivity than on relative permittivity; namely, as conductivity increases, the AL also increases [12,13].

We recall that P_AL(ℓ₁) =0.687 mW, P_AL(ℓ₂) =0.568 mW, and P_AL(ℓ₃) =0.717 mW for the male subject. When comparing the body positions ℓ₁ and ℓ₂ , the main difference is the existence/absence of the heart. At ℓ₁, there exists the heart. The heart has the dielectric properties of є_r = 58.3 and σ = 2.54 S/m. Hence, higher conductivity is experienced at ℓ₁ than at ℓ₂. In other words, more AL and higher RSS are observed at ℓ₁ than at ℓ₂. The presence of the small intestine (є_r = 54.4 and σ = 3.17 S/m) at ℓ₃ results in the highest conductivity. In summary, σ(ℓ₃) > σ(ℓ₁) > σ(ℓ₂) > where σ(ℓ₁) is the conductivity at ℓ_i. As a result, P_AL(ℓ₃) > P_AL(ℓ₁) > P_AL(ℓ₂) and P_RX(ℓ₂) > P_RX(ℓ₁) > P_RX(ℓ₃) are satisfied.

Different results are obtained for the female subject than for the male subject: P_AL(ℓ₁) > P_AL(ℓ₃) > P_AL(ℓ₂) and P_RX(ℓ₃) > P_RX(ℓ₂) > P_RX(ℓ₁). Unlike the male subject, the female has breasts, which create a small air gap (air internal of σ = 0 S/m) between the Tx antenna and the body surface. This air gap reduces the AL. The effective dielectric constant of the antenna can also vary with the air gap, which also distorts the impedance matching and radiation characteristics of the antenna. These affect RSS in addition to the body AL, and this is the main reason why P_RX(ℓ₁) and P_RX(ℓ₂) are lower than P_RX(ℓ₃) for the female. Unfortunately, it is not easy to measure the variations and to analyze their effects in the time-domain. Note that the body AL is not a function of only the dielectric properties of the human tissues. The AL is also related to the weight and size of each human tissue, their distributions, and the body structures, which will be investigated in future work. This is another reason that explains the differences in the results from the male and female subjects.

The complexity and variety of body structures may preclude the derivation of explicit formulas that compute the body AL and measure its quantitative effect on the RSS. However, our investigation showed that the body AL is a dominant parameter influencing the RSS.

For the statistical analysis, all the RSS measurements from each Tx antenna position were normalized such that normalized RSS ∈ [0,1] dB. The normalized RSS is denoted by x ∈ [0,1]. In order to perform the statistical analysis of the results, the normalized RSS distributions are compared with well-known theoretical distribution models: Normal, Weibull, Gamma, Rician, and Nakagami-m [14]. The parameter estimates of the theoretical distributions are obtained by using the maximum likelihood estimation method. The best fit distribution for each measurement is determined using the Kolmogorov-Smirnov goodness-of-fit test (KS test) [14]. This test calculates the maximum distance between the empirical distribution function (EDF) of the measured data and the cumulative distribution function (CDF) of the theoretical reference model. Among the theoretical models, the one having the minimum distance is selected as the best fit distribution.

The KS statistic is then defined as [14]

where F_E(x) is the EDF value of the measured data and F_x(x) is the CDF value of a theoretical reference distribution at x ∈ [0,1]. Table 3 presents the KS statistic values for the five distributions. The KS statistic value is a measure of the fitting error: As the KS statistic value is smaller, the reference distribution provides better fit. In our investigation, the Gamma distribution was the best fitting model. Figs. 5-7 show the EDFs of the measured data and the best fit Gamma distributions for the male and female subjects when the Tx antenna is mounted on ℓ₁, ℓ₂, and ℓ₃.

Meanwhile, the Gamma probability density function is given by

where a is the fading parameter, b is the scale parameter, and Γ(a) is the Gamma function; a and b are referred to as the Gamma parameters. The estimated parameters (a, b) of the Gamma distributions for the male and female subjects at each Tx antenna position are shown in Table 3. We denote the estimates of a and b at ℓ_i by (a, b|ℓ_i). For the male subject, (a, b|ℓ₁) = (6.4765, 0.0719) at ℓ₁ , (a, b|ℓ₂) =(7.2015, 0.0709) at ℓ₂, and (a, b|ℓ₃) =(7.9090, 0.0640) at ℓ₃. For the female subject with respect to the Tx antenna positions, (a, b|ℓ₁) = at (7.6039, 0.0666) ℓ₁ , (a, b|ℓ₂) =(7.2479, 0.0706) at ℓ₂, and (a, b|ℓ₃) = (13.190, 0.0436) at ℓ₃.

This paper investigates the on-to-off body channel characteristics at 2.45 GHz with respect to different Tx antenna positions. The experiments and simulations confirmed that different channel characteristics are observed according to the on-body Tx antenna positions and that they are dependent on the AL generated by the human body. Moreover, due to the differences in the body morphology and structure, we observed that the male and female subjects have considerably different channel characteristics. Statistical analyses were performed by comparing the measurement data to the theoretical reference distribution models. For all measurement sets, the Gamma distribution function provided the best fit to the experimental data. Our experimental results can serve as a good reference for future work on the on-tooff body channel characteristics, taking into account the effects of the antenna positions and the body AL. However, these results are presented only as guidelines since human beings have different body sizes, body structures, and body shapes, which need to be further investigated.