Multi-Frequency Wireless Channel Measurements and Characteristics Analysis in Indoor Corridor Scenarios

This multi-frequency channel measurement campaign was conducted by Keysight time domain channel sounder illustrated in Fig. 1. This system can fully support sub-6 GHz indoor channel measurement with signal bandwidth of 320 MHz. Table I lists a detailed configuration and parameters of this measurement.

This channel measurement campaign was carried out in a typical indoor corridor scenario with 41 m long, as shown in Fig. 2. Both Tx and Rx antennas were placed on the trolley, and the height of Tx and Rx antennas are 1.95 m and 1.45 m, respectively. The corridor was flanked by wooden doors, glass walls, and concrete walls. The layout of this scenario and the locations of Tx and Rx are illustrated in Fig. 3. Tx1 and Tx2 represent the position of Tx in LOS and NLOS scenarios, respectively, and the red dots represent the position of Rx. In the LOS scenario measurement, Rx was moved from position 1 to 37, from position 1 to 21 with an interval of 0.8 m between each two positions, and from position 21 to 37 with an interval of 1.6 m between each two positions. While in the NLOS scenario measurement, Rx was moved from position 3 to position 37. We measured each position 5 times, and each time with 400 snapshots. During this measurement, both Tx and Rx remain stationary.

The closed-in (CI) free space reference PL model, FI PL model and free space PL model are three commonly used PL models. The CI PL model is

where $f$ is the center frequency, $n$ is the PLE, $d$ is the distance between Tx and Rx , and $X_{\sigma}$ is the shadowing fading (SF), which can be modeled by a normal distribution of $N(0,\sigma)$, $\sigma$ represents the standard variance of SF (dB).

The FI PL model is given by

where $\alpha$ represents the PLE and $\beta$ represents the path loss offset determined from the measured data.

The free space path loss model is

In order to reduce the impact of noise, CIR with signal to noise ratio (SNR) less than 25 dB are eliminated, the remaining CIRs of 5 measurements at each position are averaged to obtain the APDP, which is calculated by

where $h_{n}(\tau)$ represents the averaged CIR of the $n$th measurement, $N$ is the total number of measurement times, here is 5, and $|\cdot|$ stands for absolute value.

The peak search algorithm is used to extract multipath components (MPCs) based on the APDP which has eliminated the effect of small scale fading. The larger value of noise floor plus 6 dB [13] and maximum power minus 25 dB is selected as the multipath power decision threshold. The received power is calculated by summing the power of all MPCs

where $P_{l}$ represents the power of the $l\text{th}$ path and $L$ represents the total number of MPCs. Then the PL is calculated by

where $P_{r}$ and $P_{thr}$ are the received power of channel measurement and system calibration, respectively. $P_{t}$ and $P_{tht}$ represent the transmitted power of channel measurement and system calibration, respectively. $G_{r}$ and $G_{t}$ are the gains of Tx and Rx antennas, respectively, both of them are 2 dBi, and $G_{att}$ represents the attenuation value of attenuator during system calibration.

The root mean square (RMS) DS is defined as

where $\tau_{l}$ represents the delay of the $l\text{th}$ paths.

The frequency domain moment estimation method of broadband channel proposed in [15] is used to estimate KF, which is suitable for the KF estimation in the static case. The frequency domain moment estimation method is calculated by

where $H_{i}$ is the response of the $i$-th frequency point, $I$ is the number of narrowband spectral samples, $G_{a}$ and $G_{v}$ represent the first and second moments of the spectral samples respectively, and $\widehat{K}$ represents the estimated KF.

The results of APDPs in LOS scenario at three frequency bands are shown in Fig. 4. The delay of LOS path at all measurement positions is shifted to the same value, and the data points above 6 dB above noise floor are detected to ensure the image can show the effective multipath more clearly.

It can be seen from Fig. 4 that APDPs of the three frequency bands have obvious reflection path, and the delay of this reflection path decreases with the increase of distance between Tx and Rx, indicating that the path comes from the end of corridor. The reflection paths of 2.4, 5 and 6 GHz are first seen at positions 4, 9 and 17, respectively, and the distance of reflection path first seen is getting shorter and shorter, which means that the attenuation of reflection path increases with increasing frequency. In addition, it can be seen that the power of LOS path decreases with increasing the distance from Tx between Rx.

The measurement results of PL at 2.4, 5, and 6 GHz frequency bands in LOS and NLOS scenarios are shown in Fig. 5. The detailed parameters of CI and FI models are listed in Table II.

It can be seen from Fig. 5 that the prediction accuracy of FI model in LOS scenario is similar to CI model, however it is better in NLOS scenario. In NLOS scenario, the CI model will underestimate the PL when the Tx and Rx are close to each other and the CI model will overestimate the PL when the distance between Tx and Rx is far.

In addition, the PL at 2.4 GHz is smaller than free space PL in LOS scenario, however the PL of 5 and 6 GHz is basically the same as free space PL, this may be caused by the stronger waveguide effect at 2.4 GHz. The PL in NLOS scenario is larger than that in LOS scenario. Both in LOS and NLOS scenarios, the PL at 2.4 GHz is the smallest, and the PL at 5 and 6 GHz is close. By observing Table II, it can be found that the PLE and $\beta$ value in the FI model tend to increase with the increase of frequency, which is consistent with [6].

The RMS DS in LOS and NOLS scenarios is listed in Table III. It can be found that the mean values of DS at three frequency bands are close to each other in the LOS scenario, and there is no obvious frequency dependence. This is in line with the conclusion in [6].

Comparing DS in LOS and NLOS scenarios, we can find that the standard deviation (STD) values of DS in NLOS scenario are smaller than that in LOS scenario at three frequency bands, and the mean values of DS in LOS and NLOS scenarios at 2.4 and 5 GHz are basically the same. This may be due to the existence of a strong reflection path in the LOS scenario. The delay of this path is much larger than that of LOS path, and the power is comparable to LOS path, increasing the mean value of DS in LOS scenario. Moreover, this reflection path does not exist at all positions, which increases the STD value of DS in LOS scenario.

Fig. 6 shows the cumulative distribution function (CDF) of KF at three frequency bands in LOS scenario, which can be fitted well by normal distribution.

As can be seen from Fig. 6 that the KF values of the three frequency bands are very close, which is consistent with the changing trend of DS. This may be because the environment does not change significantly during the channel measurement. A summary of detailed parameters description is listed in table III.