Prototyping and Experimental Results for Environment-Aware Millimeter Wave Beam Alignment via Channel Knowledge Map

The NI USRP-2974 is a Software Defined Radio (SDR) stand-alone device built with a Field-Programmable Gate Array (FPGA) and an onboard processor for rapidly prototyping high-performance wireless communication systems. With a real time bandwidth of up to 160MHz, two daughter boards covering $10$MHz to $6$GHz are utilized to perform analog up/down conversion. The baseband signal sampling/interpolation and digital up/down conversion are realized rapidly on the Xilinx Kintex-7 410T FPGA, while the Intel i7 onboard processor is responsible for waveform related operations like modulation/demodulation. Configured by the NI LabView Communication program, USRP-2974 can be operated as a transmitter to generate required signals flexibly and a receiver to process signals efficiently.

The mmPSA TR16-1909 is used as a phased antenna array for mmWave communication systems, which is a 16-element phased array and provides vertical polarization service. mmPSA TR16-1909 is cable-connected to USRP, to which the control frame is transmitted from the NI LabView Communications program to realize exhaustive beam sweeping. There are two control frame transmission methods, namely the SPI interface via HDMI line and the RS232 interface via USB-RJ45 connection line.

The device supports time division duplex (TDD) communication, whose operating frequency band is 27-29 GHz. The phase array provides the fast and normal modes in phase control. In the fast mode, 16 phase modules can be set in dependently, and the setting accuracy of each phase module is 5 bits. And in the normal mode, 16 phase modules are set simultaneously, and the setting accuracy is increased by 1 bit compared to the fast mode. The mmPSA TR16-1909 has a codebook with a scale of 64, where the angular range is from $-56^{\circ}$ to $+56^{\circ}$ in the horizontal direction.

The UWB indoor positioning system is used to measure the location of the receiver and transmitter in our experiment. The UWB technology can obtain Gbit/s data rate by transmitting nanosecond or shorter pulses. Compared with the traditional narrowband system, the UWB system has superior anti-interference performance, a high transmission rate, an extremely wide bandwidth, a large system capacity, low transmission power, good confidentiality, etc. Therefore, it has obvious advantages in short-range high-speed data transmission and high-precision indoor positioning. The Nooploop UWB positioning system used in the CKM-based beam alignment prototype has a positioning update rate of 200Hz and a positioning accuracy of 10cm.

The UWB indoor positioning system is composed of the equipment layer, solution layer, and application layer, mainly including UWB base station, UWB tag, Power Over Ethernet (POE) switch, positioning engine, LabView program platform, etc. The main workflow is as follows: UWB tags are attached to the transmitter and receiver, sending the measurement signals to the base stations in the scenario through the UWB channel. In this experiment, five positioning base stations are set up, which collect tag data and transmit it to the solution engine in order to compute the UWB tag coordinates. After that, the parsed location information data is forwarded and uploaded to the LabView host computer and can be visualized in the LabView program platform, so as to control the mmPSA and perform CKM-based beam alignment based on the location information.

The gyroscope module can efficiently obtain the real-time motion attitude with an attitude measurement accuracy of $0.2^{\circ}$ and extremely high stability, which adopts a high-performance microprocessor and advanced dynamic filtering algorithm. It can provide real-time data with an update rate of up to 200Hz, so as to meet the needs of various high-precision applications and accomplish accurate motion capture and attitude estimation.

In order to use environment information to help beam alignment, we fix the transmitter location and build a CKM for the area of interest for the receiver. In Fig. 7b, the CKM is built on a square area of size $4m\times 4m$, while in Fig. 8, the CKM is built on a square area of size $3.2m\times 4.8m$. The grid size is $0.8m\times 0.8m$. For each grid in CKM, the center coordinate is used as the grid’s coordinate. In the offline construction of the CKM, the exhaustive beam sweeping method is used to determine the optimal beam index pairs. During the construction of CKMs, the receiver orientation is set to $\omega[t]=0$. Considering that the construction steps of CKM are universal to LoS and NLoS scenarios, we take the NLoS scenario as an example and the specific process is as follows: First, the position of the transmitter is fixed, and the grids within the entire CKM are numbered from 1 to 24 in Fig. 8b. Then, the UWB tag is placed on the receiver, and the receiver is sequentially moved from grid 1 to grid 24 in sequence to find the optimal transmit and receive beam index pairs for the grid.

When the receiver moves to the center of grid $i$, the coordinate $\big{(}x_{i},y_{i}\big{)}$ of the receiver is obtained by the UWB system and stored in the CKM. After that, the beam index pair of the transmitter and receiver is switched to perform exhaustive beam sweeping. The difference between the diversified beam index pairs is reflected in the power of the signal received at the receiver. The higher the power of the received signal, the better the corresponding beam index pair. For each grid, communication tests with 64x64 beam index pairs are conducted, since each mmPSA has 64 preset beams in the codebook. The communication test for each beam pair lasts 30$\mathrm{ms}$, during which the received signal power is read three times in 10$\mathrm{ms}$ intervals and the maximum received signal power is taken to reduce the impact of beam index pair switching on the communication performance. The transmit and receive beam index pair with the maximum received power will be selected and stored in the CKM as the optimal transmit and receive beam index pair $\big{(}m^{\mathrm{t}}_{i},m^{\mathrm{r}}_{i}\big{)}$ of the reflective NLoS link for grid $i$. Therefore, the corresponding CKM for grid $i$ is recorded and saved in the following form as

The complete CKM construction process of the CKM-based beam alignment is summarized in Algorithm 2. The construction step of the CKM does not change even in complex wireless environments, i.e., sweeping the transmit and receive beam pairs in the codebook. Therefore, when the number $s_{g}$ of grids on the sides of a square area increases, which means the size of the environment increases, the construction complexity of the CKM is only $\mathcal{O}(s_{c}s_{g}^{2})$, where $s_{c}$ is the size of the codebook. This suggests that the complexity of construction does not depend on the complexity of the wireless environment.

In the static scenario with LoS link, the experiment with an arbitrary grid $i$ in the CKM is conducted. The coordinates of the receiver are obtained by the UWB positioning system, while the orientation $\omega[t]$ of the receiver is obtained by a gyroscope fixed on the receiver. The geometric representation of the CKM-based strategy in LoS scenario is also shown in Fig. 6. With continuously changing receiver orientation, the CKM-based beam alignment strategy can predict the transmit and receive beam pairs based on channel knowledge and receiver orientation. Specifically, the coordinate $\hat{\mathbf{q}}[t]=[\hat{q}_{x}[t],\hat{q}_{y}[t]]$ of the receiver can be obtained in real time through the UWB positioning system. By calculating the distance between the receiver and the center of each grid, we can find out which grid center is nearest to the real-time position of the receiver and thus determine which grid $i^{\star}$ the receiver is located in real time as

Then, the transmit beam index $m^{\mathrm{t}}_{i^{\star}}$ of grid $i^{\star}$ is obtained according to the constructed CKM, and the receive beam index is refined according to the receiver orientation $\omega[t]$ as

Fig. 9 shows the performance comparison of the two beam alignment strategies for different receiver orientations. The receiver orientation $\omega[t]$ ranges from -$40^{\circ}$ to $40^{\circ}$ at the step of $5^{\circ}$. As can be seen from Fig. 9, for different receiver orientations, the CKM-based beam alignment strategy refines the beam index pairs read from the CKM according to orientation $\omega[t]$, and the received power obtained is very close to that of the location-based strategy. Therefore, the CKM-based beam alignment strategy is able to maintain good communication performance regardless of the orientation of the receiver. This is because the CKM-based strategy exploits the channel knowledge to first obtain the original transmit and receive beam index pairs stored in the CKM. Then the beam index pair is refined according to the receiver orientation to achieve good beam alignment. Therefore, in quasi-static scenarios with LoS links, the CKM-based beam alignment strategy can obtain comparable communication performance to the location-based benchmark strategy. The performance difference between the two strategies mainly comes from the location error and the gyroscope inertia error in different receiver orientations. The CKM-based strategy has smaller variation mainly because the location errors have a large impact on the location-based strategy, but generally do not interfere with the judgment of the nearest grid center, and therefore interfere less with the CKM-based strategy.

The relevant information of the communication environment is embedded in the CKM $\{c_{i}\}$ and is expressed by the optimal beam index pair for each grid. During the experiment, the receiver can move freely within the constructed CKM area, while the prototype performs beam alignment through the already constructed CKM. For convenience, the receiver orientation is set to $\omega[t]=0$. It is worth mentioning that the receiver orientation $\omega[t]=0$ is assumed in order to demonstrate the beam alignment performance of the CKM-based strategy through NLoS links in different grids under uniform conditions. When $\omega[t]\neq 0$, the receive beam index read from the CKM should be refined according to $\omega[t]$, which is consistent with (12) in the LoS scenario. Specifically, the grid $i^{\star}$ to which the receiver belongs is first determined by the Euclidean distance of the receiver from each grid with (11). Similarly, the transmit beam index $m^{\mathrm{t}}_{i^{\star}}$ is obtained from the constructed CKM. Considering $\omega[t]=0$ in (12), the index $m^{\mathrm{r}}_{i^{\star}}$ of grid $i^{\star}$ can be obtained directly from the CKM as the receiver beam index. A similar beam index detection process is added to the CKM-based beam alignment, i.e., the control frames are retransmitted only when the transmit or receive beam index changes.

The actual scenario and the illustrative diagram of the CKM-based beam alignment prototype are respectively shown in Fig. 8a and Fig. 8b. In Fig. 10, the communication performance of CKM-based beam alignment is compared with the benchmark location-based and beam sweeping strategies. As illustrated in Fig. 10, the location-based strategy gives poor performance because it can only assume the LoS links that do not exist in the considered scenario. Meanwhile, since the reflective link considering the environment information is stored in the CKM in the form of beam index pairs, the CKM-based beam alignment strategy can use the reflective link for communication even if the LoS link is blocked. By comparing the received power at each grid, the CKM-based strategy is able to achieve a communication performance comparable to that of the exhaustive beam sweeping strategy, but with almost no real-time overhead. The modest performance loss is caused primarily by the receiver’s location error relative to the grid center during movement.

The CKM in this experiment is built on a $4m\times 4m$ square area with $0.8m\times 0.8m$ grid. The process of CKM construction in a dynamic environment is similar to Algorithm 1. The main difference in the construction of CKM from Algorithm 1 is that, only the reflective NLoS link in the environment is considered, i.e., the optimal beam index pair $\big{(}m^{\mathrm{t}}_{i},m^{\mathrm{r}}_{i}\big{)}$ needs to be found and stored in CKM. However, in the construction of CKM in the dynamic environment, both LoS and NLoS links need to be considered in order to improve the ability of the prototype to maintain communication in dynamic environments. Therefore, in the exhaustive search at each grid $i$, the optimal LoS link beam index pair $\big{(}m^{\mathrm{t}}_{i},m^{\mathrm{r}}_{i}\big{)}$ and reflective NLoS link beam index pair $\big{(}l^{\mathrm{t}}_{i},l^{\mathrm{r}}_{i}\big{)}$ need to be found out in the spatial direction according to the recorded received signal power and codebook, and stored in the corresponding CKM with the following format as

The simulation of the dynamic environment and the process of CKM-based beam alignment in the corresponding dynamic environment are presented here. During the experiment, the moving vehicle with an aluminum reflective plate will serve as a significant communication environment-blocking dynamic obstacle. The receiver can communicate with the transmitter at any location within the constructed CKM area, and during the communication, we will move the moving vehicle to introduce dynamic blockage. The moving vehicle first moves toward the LoS link between transmitter and receiver, as in Fig. 11b. Considering that the LoS link is currently unimpeded, this phase is used to simulate a scenario where the dynamic environment does not affect communication temporarily. Next, we control the moving vehicle carrying an aluminum reflective plate across the LoS between the transmitter and receiver. As shown in Fig. 11c, the moving vehicle remains stationary for a period of time as it passes through the LoS link to simulate a scenario in which the LoS link is blocked. After that, the vehicle will continue to move until it is far away from the LoS link, which is used to represent the moment when the dynamic environment changes again and the LoS link recovers.

After obtaining the grid index, the key to CKM-based beam alignment in dynamic environments is determining which beam index pair in the CKM will be used for communication. Denote the real-time coordinate of the dynamic obstacle as $\hat{\mathbf{o}}[t]=[\hat{o}_{x}[t],\hat{o}_{y}[t]]$. In order to avoid the communication outage due to the dynamic obstacle, we use the distance-based judging rule to select the communication link. Specifically, the LoS link is used by default in the communication. Once the distance from the dynamic obstacle to the LoS link is less than a given threshold $\eta$, the selected communication link will switch from the LoS link to the reflective NLoS link in advance to avoid communication outage. To represent the position relation between dynamic obstacle and transmitter and receiver, we introduce the following vector according to the coordinates obtained from UWB positioning system

where $\hat{\mathbf{b}}[t]=[\hat{b}_{x}[t],\hat{b}_{y}[t]]$,$\hat{\mathbf{q}}[t]=[\hat{q}_{x}[t],\hat{q}_{y}[t]]$ and $\hat{\mathbf{o}}[t]=[\hat{o}_{x}[t],\hat{o}_{y}[t]]$ denote the real-time coordinates of the transmitter, receiver and dynamic obstacle, respectively. $\overrightarrow{\mathrm{TR}}$ represents the vector pointing from the transmitter to the receiver, while $\overrightarrow{\mathrm{TO}}$ represents the vector pointing from the transmitter to the dynamic barrier.

As shown in Fig. 12, the ratio $r$ of the length of the projection of $\overrightarrow{\mathrm{TO}}$ on $\overrightarrow{\mathrm{TR}}$ to the length of $\overrightarrow{\mathrm{TR}}$ is calculated in (16), while the distance $d_{\mathrm{o}}$ from the dynamic obstacle to the LoS link is expressed as in (17)

A complete CKM-based beam alignment process in dynamic environments is shown in Algorithm 3.

In Fig. 13, we compare the performance of the CKM-based beam alignment with the benchmark location-based beam alignment in the dynamic environment. The received power is recorded versus different distances $d_{\mathrm{o}}$ from the dynamic obstacle to the LoS link with the step $10$cm. Considering that half of the width of the aluminum reflective plate on the moving vehicle is $25$cm, the distance threshold is set as $\eta=30$cm. When the moving vehicle representing the dynamic obstacle just starts to move, it is far away from the LoS link, which is shown in Fig. 13 with $-50\mathrm{cm}\leq d_{\mathrm{o}}<-30\mathrm{cm}$. In this case, the LoS link is almost undisturbed by the dynamic obstacle, and the CKM-based beam alignment strategy chooses the LoS link with $\big{(}m^{\mathrm{t}}_{i},m^{\mathrm{r}}_{i}\big{)}$ for communication, which is consistent with the results obtained by the location-based strategy. This explains the coincidence of the two received power lines with $-50\mathrm{cm}\leq d_{\mathrm{o}}<-30\mathrm{cm}$. When $-30\mathrm{cm}\leq d_{\mathrm{o}}\leq 30\mathrm{cm}$, which means $|d_{\mathrm{o}}|<\eta$, different beam index pairs are selected for communication by the benchmark location-based and the CKM-based strategies. On the one hand, the location-based strategy has no knowledge of the dynamic environment and can only choose the beam index pair of the LoS link. Thus, as the distance $d_{\mathrm{o}}$ changes from $-30$cm to $30$cm, the LoS link is first gradually obstructed by the dynamic obstacle and then gradually recovers, which also leads to a decrease and then an increase in the received power and a corresponding minimum at $d_{\mathrm{o}}=0$cm. On the other hand, the CKM-based beam alignment strategy fully considers the impact of dynamic obstacle and switches to NLoS beam index pair $\big{(}l^{\mathrm{t}}_{i},l^{\mathrm{r}}_{i}\big{)}$ in time ($|d_{\mathrm{o}}|\leq\eta$). When the CKM-based strategy switches to the NLoS link for communication, the corresponding received power has a power loss up to 5dB, which corresponds to the path loss between the alternative NLoS link and the LoS link. Compared with the location-based strategy, the resulting gain of the received power is up to $8$dB or more. This fully demonstrates that the CKM-based beam alignment can effectively cope with the variation of environment and improve communication performance through pre-stored channel knowledge. Notice that the location-based strategy performs better at $d_{\mathrm{o}}=30$cm. This can be explained by the spatial limitations of the experimental site, where the dynamic obstacle moves closer to the copper reflector as the moving vehicle moves, thereby impacting the NLoS link and diminishing its performance to some degree. As the distance increases to $d_{\mathrm{o}}\geq 30$cm, the CKM-based strategy treats the impact of dynamic obstacle on the LoS link as negligible and switches the beam index pair back to the LoS link $\big{(}m^{\mathrm{t}}_{i},m^{\mathrm{r}}_{i}\big{)}$. Therefore, the curves of the received power are once again consistent for both the benchmark and the CKM-based beam alignment.