Initial Experimentation of a Real-Time 5G mmWave Downlink Positioning Testbed ^†

Abstract

This work presents the initial experimentation of a real-time 5G mmWave downlink positioning testbed deployed at Airbus premises. This experimentation is part of a first-of-a-kind testbed for hybrid Global Navigation Satellite Systems (GNSS), fifth-generation (5G) new radio (NR) and sensor positioning, called the Hybrid Overlay Positioning with 5G and GNSS (HOP-5G) testbed. The mmWave 5G base station (BS) exploits the 5G standard positioning reference signal (PRS) to support positioning capabilities within the 5G NR downlink transmissions. Outdoor field results are used to characterize the received power levels and beam-based angle-of-arrival (AoA) estimation accuracy of this 5G mmWave PRS platform. The goal is to assess the suitability of this platform to enhance the positioning performance thanks to the 5G downlink mmWave transmissions. To the best of the authors’ knowledge, this paper presents the first AoA results using OpenAirInterface (OAI) PRS mmWave signal transmissions at 27 GHz for positioning. These initial field results indicate a maximum coverage of 30 m and an AoA accuracy limited by the reduced array size. The limitations and potential enhancements of this platform are provided as future recommendations.

1. Introduction

Millimeter-wave (mmWave) communications at carrier frequencies above 24 GHz are one of the disruptive technologies in fifth-generation (5G) new radio (NR) systems, thanks to the large bandwidth available in those high frequency bands. In addition, the large path loss suffered at mmWave bands leads to the use of directional antennas or beamforming techniques together with a high network density, offering angle-of-arrival (AoA) capabilities under line-of-sight (LoS) conditions. As a result, these broadband transmissions with beamforming capabilities in favorable propagation conditions also offer a key disruptive feature for 5G precise positioning targeting submeter-level accuracies [1,2], as well as for sixth-generation (6G) localization and sensing use cases [3].

Dedicated 5G network features for positioning purposes are already specified in the 3GPP NR standard, such as the 5G-based positioning methods in [4], the NR positioning reference signal (PRS) in [5], and the NR positioning protocol A (NRPPa) in [6]. Indeed, the achievable positioning capabilities of 5G-based positioning and its enhancements, such as with downlink time–difference of arrival (DL-TDoA), have been evaluated in the 3GPP standardization since Release 16 in [7], through simulations for frequency range 1 (FR1) below 6 GHz and for FR2 above 24 GHz. However, the real-world demonstration of 5G precise positioning has been very limited, due to major challenges when developing a real-time 5G network able to exploit the high bandwidths required for precise positioning. For instance, an FR1 field trial of 5G positioning based on single- and multi-cell round-trip time (RTT) and uplink angle-of-arrival (AoA) was demonstrated in [8] to achieve meter-level accuracies in FR1 LoS conditions. Another relevant example is the Hybrid Overlay Positioning with 5G and GNSS (HOP-5G) testbed [9,10,11,12], where the world’s first airborne 5G network for positioning purposes was shown to enable submeter localization accuracy with 80-MHz downlink PRS in FR1 LoS. In terms of prototype 5G FR2 network deployments, 5G mmWave precise positioning was demonstrated to achieve meter-level accuracies with downlink angle-of-departure (AoD) positioning in [13], or submeter-level accuracies with DL-TDoA in [14], depending on the network synchronization. Nonetheless, there is very limited literature on 5G mmWave network deployments for precise positioning in outdoor environments.

There are certainly extensive efforts in the literature to develop high-end mmWave testbeds able to exploit the most advanced 5G FR2 and future 6G communication features. State-of-the-art examples of mmWave front ends and testbeds at 28 GHz and 60 GHz are summarized in [15], including an overview of the example COSMOS testbed able to operate in both mmWave bands. Recently, an end-to-end connection between the OpenAirInterface (OAI) FR2 network and a commercial UE has been demonstrated in [16]. High-end mmWave front ends with a 64-element antenna array for a single 5G transmitter and a 256-element array for a single 5G receiver are deployed in [17], over a controlled indoor LoS environment to achieve centimeter-level ToA accuracy. Field results with a 5G mmWave base station and a prototype UE with 16-element array are shown in [18], to assess experimental RTT, downlink AoD and uplink AoD positioning for vehicular applications.

This work is a continuation of the FR1 positioning contributions within the HOP-5G testbed in [9,10,11,12]. In contrast to prototype commercial network demonstrators, the HOP-5G testbed is an experimental and flexible proof-of-concept (PoC) of advanced 5G and hybrid positioning features, based on OAI open-source software, commercial off-the-shelf (COTS) computing and software-defined radio (SDR) equipment, as well as custom-built payload platforms for unmanned aerial vehicles (UAVs) or drones in aerial networks. In this paper, a 16-element mmWave front end at both the transmitter and receiver is presented to assess the real-time beam-based AoA performance with downlink 5G PRS transmissions. To the best of the authors’ knowledge, this is the first work on a real-time 5G mmWave PRS testbed with reduced antenna arrays assessing downlink AoA performance outdoors.

2. OAI-Based 5G mmWave Network

OAI is an open-source software that implements a full 5G protocol and is widely used in research and experimental testbeds [19]. During recent years, OAI has also introduced the support of positioning capabilities, with key developments originated from the HOP-5G project. The physical layer includes the sounding reference signal (SRS) as in 3GPP Release 15 and the PRS as in 3GPP Release 16, both necessary to perform UL- and DL-based positioning methods, respectively. Moreover, OAI has implemented the NRPPa protocols and procedures to perform standard-compliant UL-TDoA positioning [20], where a development of the Location Management Function (LMF) also exists.

There are several examples on the use of OAI for positioning, in addition to the HOP-5G testbed. In [21], OAI is combined with a multi-channel SDR and a custom 2 × 2 patch antenna array for enhanced cell-ID (E-CID) positioning in indoor scenarios, by using RTT and AoA estimates based on the demodulation reference signal (DM-RS). Thanks to the OAI PRS developments in the HOP-5G project, a testbed with four OAI base stations or gNBs, and the OAI user equipment was deployed in [22] to exploit the PRS transmissions for DL-TDoA positioning. More recently, a novel framework was presented in [23] to estimate the RTT between a UE and a gNB with existing 5G NR transmissions.

In this paper, the OAI-based 5G mmWave network is based on downlink PRS transmissions between a single gNB and a single UE to estimate the time-of-arrival (ToA) and the received signal reference power (RSRP) at the OAI UE. The following subsections further describe the ToA and RSRP estimators.

2.1. Positioning Reference Signal

The 5G NR PRS is specified in the 3GPP standard since Release 16 as a reference signal dedicated for positioning purposes, with enhancements introduced as of the current standard in [5]. The PRS can be flexibly configured to adapt to the positioning application, by using tailored configurations of the PRS sequence and its resource allocation, such as based on the number of consecutive orthogonal frequency-division multiplexing (OFDM) symbols per slot, bandwidth or beams. In OAI, the PRS is explicitly implemented for 30 kHz and 120 kHz subcarrier spacing, corresponding to numerology 1 and 3, respectively. Furthermore, the implementation of PRS in OAI also supports the flexible resource allocation in terms of frequency, time and beams.

The resource allocation of the PRS implementation in OAI mmWave configuration is here defined by a subcarrier spacing of 120 kHz, 64 physical resource blocks (PRBs) occupying a bandwidth of 92.16 MHz, 6 consecutive OFDM symbols, comb-size of 4 (i.e., one every four subcarriers used for PRS resource elements), and allocated for 8 different beams. The resulting PRS resource allocation within the 5G NR radioframe and within every PRS slot is shown in Figure 1.

2.2. Time-of-Arrival and Power Measurements

The ToA estimation is here based on a threshold-based correlation of the PRS, i.e., ToA for the maximum peak of the PRS correlation above a certain threshold. This is implemented with a least-squares channel estimate in the frequency domain from the PRS, which is then converted to the time domain using an Inverse Discrete Fourier Transform (IDFT) to determine the ToA.

The least-squares estimate of the PRS channel $h\left[k\right]$ at the k-th subcarrier can be obtained from the frequency–domain received signal $y\left[k\right]$ as

$$h\left[k\right]=y\left[k\right]\cdot s^{\ast }\left[k\right],$$

(1)

where k ∈ [0, K – 1] and K is the total number of subcarriers per PRS symbol. The frequency domain channel estimates can be represented in vector form as

$$\mathit{h}=\left[h\left[0\right],h\left[1\right],\dots ,h[K-1]\right]\in {\mathbb{C}}^{K\times 1}.$$

(2)

Then, $\mathit{h}$ is zero-padded by an up-sampling factor $K_{\mathrm{n}}$ represented as ${\mathit{h}}_{\mathrm{Z}\mathrm{P}}\in {\mathbb{C}}^{K_{\mathrm{n}}\cdot K\times 1}$ to increase the time resolution when converted to the time domain. The ToA estimate from ${\mathit{h}}_{\mathrm{Z}\mathrm{P}}$ is finally obtained by taking an IDFT of size ${K\cdot K}_{\mathrm{n}}$ as

$$\widehat{\tau }={\displaystyle \frac{1}{F_{\mathrm{s}}\cdot K_n}}\cdot \mathrm{a}\mathrm{r}\mathrm{g}\mathrm{m}\mathrm{a}\mathrm{x}\mathrm{I}\mathrm{D}\mathrm{F}\mathrm{T}\left\{{\mathit{h}}_{\mathrm{Z}\mathrm{P}}\right\},$$

(3)

where $F_{\mathrm{s}}$ is the sampling rate.

The RSRP in the digital domain (measured in dBFS) is estimated from the received signal $y\left[k\right]$ (which is in Q1.15 fixed-point format) as follows:

$$\mathrm{R}\mathrm{S}\mathrm{R}\mathrm{P}[\mathrm{d}\mathrm{B}\mathrm{F}\mathrm{S}]=10\cdot {\log }_{10}\left({\displaystyle \frac{1}{K}}\sum _{k=0}^{K-1}{\left|y\left[k\right]\right|}^2\right)-10\cdot {\log }_{10}{2}^{30},$$

(4)

where $\left|\cdot \right|$ denotes a modulus of a complex number and $2^{30}$ corresponds to the power level at full scale of 30 bits. Furthermore, to convert RSRP from dBFS to dBm, the RSRP in dBm is obtained by

$$\mathrm{RSRP}\left[\mathrm{d}\mathrm{B}\mathrm{m}\right]=\mathrm{RSRP}\left[\mathrm{d}\mathrm{B}\mathrm{F}\mathrm{S}\right]-G_{\mathrm{R}\mathrm{X}},$$

(5)

where $G_{\mathrm{R}\mathrm{X}}$ is the receiver gain, which is calibrated using a signal at a known input level.

3. Beam-Based Angle-of-Arrival Algorithm

The AoA estimation is here performed with a beam-based AoA algorithm. First, the the mmWave front end at the receiver scans a certain angle estimation range by sweeping the pre-selected analog beams from the beambook. Then, the AoA is estimated by weighting the received power measurements of each scanned analog beam.

A polynomial fitted to the RSRP measurements of the PRS allows us to interpolate the signal strength between the receive beams and estimate the AoA of the signal with accuracy exceeding the beam separation. Let us assume that the UE receives the PRS signal via $M$ receive beams, with beam directions in the azimuth angle domain $\mathsf{\varphi }\in {\mathbb{R}}^{\mathrm{M}}$ spanning the desired AoA estimation range that includes the (unknown) AoA of the LoS path, ${\varphi }_{AoA}$. Let us also assume that the signal received via each receive beam is in a form of raw I/Q samples, with I and Q signals’ output recorded simultaneously through the separate output ports to obtain the baseband real part and imaginary part of the received signal, respectively. Assuming appropriate signal synchronization and cyclic prefix (CP) removal, the output of these two ports can be combined into a vector of complex samples as ${\mathit{S}}_{\mathrm{R}\mathrm{X}}\in {\mathbb{C}}^{\mathrm{N}}$, where $N$ is the number of OFDM symbol samples. Each beam-based RSRP measurement, therefore, can be obtained as

$$P_{\mathrm{b}}=10\cdot {\log }_{10}\left({\displaystyle \frac{{\sum }_{i\in \mathsf{\Omega }}{\left|{\mathit{X}}_{\mathrm{R}\mathrm{X},i}\right|}^2}{K}}\right)\in \mathbb{R},$$

(6)

where ${\mathit{X}}_{\mathrm{R}\mathrm{X},i}$ denotes the discrete Fourier transform of ${\mathit{S}}_{\mathrm{R}\mathrm{X}}$, and $\mathsf{\Omega }$ denotes the set of PRS active subcarrier indices of cardinality $K$. The RSRP measurements for all RX beams can be represented as a function of the beam azimuth angle ${\mathbf{P}}_b\left(\mathsf{\varphi }\right)\in {\mathbb{R}}^{\mathrm{M}}$, corresponding to the horizontal RSRP profile of the received signal.

The proposed AoA estimation algorithm is then built upon a polynomial fitting of the RSRP profile using the weighted least squares (WLS) method. In particular, the vector ${\mathbf{P}}_b$ is approximated with the second-degree polynomial (parabola) $\widetilde{{\mathit{P}}_b}=c_0+c_1\varphi +c_2{\varphi }^2$ subject to a Gaussian weighting function $w\left(\varphi \right)=e^{-{\displaystyle \frac{1}{2}}{\left({\displaystyle \frac{\varphi -{\varphi }_{\mathrm{m}\mathrm{a}\mathrm{x}}}{\sigma }}\right)}^2}$, which is centered around the beam azimuth with the largest power ${\varphi }_{\mathrm{m}\mathrm{a}\mathrm{x}}=\mathrm{a}\mathrm{r}\mathrm{g}\mathrm{m}\mathrm{a}\mathrm{x}\left({\mathit{P}}_b\right)$ and standard deviation $\sigma$, being a user-defined parameter. The AoA estimate can be calculated in closed form as a function of the polynomial coefficients:

$${\widehat{\varphi }}_{\mathrm{A}\mathrm{o}\mathrm{A}}=-c_1/\left(2c_2\right).$$

(7)

In case the RSRP measurements suffer from errors due to very low SNR or interference, or in case the amount of receive beams is not enough to define the maximum of ${\mathbf{P}}_b$, the parabola may be inverted or degenerated, which does not allow to estimate ${\varphi }_{AoA}$. A condition $c_2<0$ ensures that this does not happen, and AoA can be estimated using Equation (7). Note that in case this condition does not hold and parabolic approximation does not exist, AoA can be estimated as a direction of the beam with highest RSRP measurement, i.e., ${\widehat{\varphi }}_{\mathrm{A}\mathrm{o}\mathrm{A}}={\varphi }_{\mathrm{m}\mathrm{a}\mathrm{x}}$.

An example result of the AoA beam-based algorithm is shown in Figure 2.

Note that the above beam-based AoA algorithm is presented for the azimuth beamforming at the UE. However, in case of elevation beamforming, the same principle can be applied to the elevation or zenith angle ${\theta }_{\mathrm{A}\mathrm{o}\mathrm{A}}$ using the vertical RSRP profile or extended to fitting a parabolic surface in case of tridimensional (3D) beams.

4. HOP-5G mmWave Positioning Testbed

The HOP-5G mmWave positioning testbed is based on a fixed mmWave gNB and moving mmWave UE, as it is shown in Figure 3. This testbed follows a modular architecture to enable the operation of each node as transmitter or receiver with specific hardware and software configurations. The key hardware components of the testbed are as follows:

• mmWave front end: The Sivers EVK02001 is an evaluation kit with an analog beamformer mmWave front end (FE) able to up-convert 5G NR signals from baseband to mmWave or to down-convert from mmWave to baseband, by integrating the Sivers TRXBF02 COTS transceiver. The 16-element antenna array, i.e., non-commercial uniform rectangular array (URA) with 2 × 8 elements, enables steerable beams within the beambook upon operator commands.
• Software-defined radio: The USRP X310 is a flexible SDR, which here includes a BasicTX or BasicRX daughterboard for transmission or for reception, respectively, and a GPS disciplined oscillator (GPSDO) to increase the stability of the SDR clock. This SDR is used with a sampling rate of 92.16 Msps to stream the baseband samples of the 5G NR signals between the mmWave front end and the host computer.
• Host computer: The host computer includes the necessary software modules to establish 5G transmission, 5G reception, GNSS data collection and AoA estimation. At the transmission, the OAI gNB is used to stream the 5G PRS to the mmWave front end through the SDR in real time. At the reception, the OAI UE is used to process the 5G PRS and to obtain ToA and RSRP measurements in real time, and the AoA estimator controls the beams of the mmWave front end and computes the AoA measurements also in real time. In both nodes, the GNSS receiver manager operates and collects the GNSS position solution from the receiver in real time.
• GNSS receiver: The u-blox F9P is a multi-band high precision GNSS receiver, whose position solution is used for the ground truth of the field campaign.
• GNSS antenna: A survey multi-band GNSS antenna is used at both nodes together with a splitter to feed GNSS signals to the USRP GPSDO and to the GNSS receiver.

The AoA estimation software module at the UE host computer controls the mmWave front end to enable real-time AoA measurements. This module connects to the Sivers EVK and initiates its operation. Then, the software proceeds to establish the reception beam on the antenna array. As it is described in Section 3, beam-based AoA estimation is performed with the RSRP measurements of the PRS received from the gNB for each beam. Thus, the AoA algorithm collects RSRP measurements for each beam angle and proceeds to sweep each beam within −45 degrees to 45 degrees with a 4.5-degree resolution per beam. The reception duration for each beam angle is 20 ms, which is sufficient for the OAI UE to perform and disseminate the PRS RSRP and ToA measurements. The measurement dissemination is performed over the MQTT protocol. OAI publishes the measurements messages in a specific MQTT topic, and the AoA algorithm subscribes to this topic to collect the measured RSRP values for each reception beam angle. Since the RSRP collection lasts 20 ms for each beam, the overall measurement collection has a duration of 420 ms for all beams. After the RSRP collection, the AoA algorithm estimates the reception angle in real time. The AoA algorithm also collects the most recent position solution published by the GNSS receiver manager over MQTT. Finally, the AoA algorithm outputs the AoA estimation together with the RSRP measurements and GNSS position solution.

5. Field Campaign

This section discusses the initial experimentation results on the maximum coverage distance and receiver orientation impact, when using the HOP-5G mmWave positioning testbed in a field campaign with LoS conditions.

5.1. Description

The HOP-5G mmWave positioning testbed was deployed on demand at Airbus premises for experimentation purposes. The FR2 downlink transmission was established at 27 GHz with a static OAI gNB and a moving OAI UE in LoS. The 5G PRS configuration described in Section 2.1 was used to achieve a bandwidth of 92.16 MHz. As it is shown in Figure 4, the starting position of the UE was as close as possible to the gNB, and then the UE (mounted on a trolley) was moved linearly until a maximum coverage distance. The UE moved with a nearly linear trajectory at a nearly constant and low velocity, i.e., with variations between 0.25 m/s and 1 m/s, except at planned temporary stops to evaluate the static scenario.

The field experiment was based on a five-step procedure repeated in each test to perform the AoA measurements:

• The OAI UE acquires and synchronizes to the 5G NR signals transmitted from the OAI gNB.
• The GNSS receiver manager collects and disseminates the GNSS-based UE position solution in real time.
• For each beam within [−45°, −41.5°, …, 41.5°, 45°], the AoA algorithm enables the corresponding beam on the Sivers EVK, the OAI UE estimates the RSRP from the PRS subframes, and the AoA algorithm collects the RSRP measurement for the specific beam.
• The AoA algorithm performs real-time AoA estimation from the collected RSRP measurements.
• The AoA estimate is disseminated together with the latest GNSS position solution.

The data measurements with the AoA, RSRP and GNSS solution were logged in real time and post-processed for performance analysis.

5.2. Maximum Coverage Distance

The assessment of the maximum coverage distance was performed based on the availability of OAI measurements as the UE moved farther from the gNB. The distance between gNB and UE was computed with the stand-alone GNSS solution estimated at both nodes during the entire duration of the field test (i.e., 120 s).

The field results shown in Figure 5 indicate a maximum distance around 30 m between the UE and gNB with the HOP-5G mmWave positioning testbed. At 30 m of distance, the OAI UE lost the 5G NR signal synchronization, and the test was stopped before the UE returned to the starting point for the following test. The AoA estimation accuracy was analyzed based on the RSRP distribution for each beam as a function of the distance between the UE and gNB, as it is shown in Figure 5a,b. Considering the static periods within the trajectory, the average RSRP was computed for each beam angle. The result indicates an almost constant average RSRP value between beam angles −13.5° and 13.5°, while the rest of the beam angles (outside of this range) include values below 3 dB. This beam pattern was kept over the different distances. As a result, when both the transmitter and receiver are aligned, i.e., an azimuth orientation of 0°, the AoA estimation accuracy is equal to 4.6° at static periods. The AoA performance degrades to 9.5° for the full trajectory, which includes dynamic and static periods. Certainly, the reduced array size limits the AoA estimation accuracy.

5.3. Impact of the Receiver Orientation

The receiver azimuth orientation also has a significant impact on the beam alignment between the transmitter and the receiver. In order to assess this impact on the FR2 AoA estimation performance, several tests were repeated for different receiver orientations compared to the perfect alignment studied in the previous section, i.e., with UE orientations at 0°, 30° and −30°. The static periods of these tests were used to obtain the beam pattern for each orientation. As it is shown in Figure 5c, the beam pattern of the Sivers EVK is not symmetric, and there are abrupt changes in the RSRP between −13.5° to −18° and between 13.5° and 18°, which correspond to angle offsets of 45° with respect to the antenna boresight. This reduces the sensitivity of the AoA to small changes in the receiver orientation, especially when perfectly aligned. Future work can explore enhanced AoA weighting functions optimized for the beam response of the antenna array or based on iterative fitting, as in [24].

6. Conclusions

This paper presents initial results of a real-time 5G mmWave testbed, able to exploit positioning reference signal (PRS) transmissions for beam-based angle-of-arrival (AoA). The testbed is based on the combination of OpenAirInterface (OAI), a software-defined radio (SDR) and a 16-element mmWave front end to be deployed at both the base station (BS) and user equipment (UE) and operating at 27 GHz. The proposed beam-based AoA algorithm controls the antenna array to sweep beams over the scanned angles, while collecting receiver power measurements for each beam. The initial experimentation results of the field campaign indicate a maximum distance coverage of 30 m between the BS and UE of the testbed. In addition, an AoA estimation accuracy of 4.6° is achieved under UE static conditions and perfect alignment between transmitter and receiver arrays. The introduction of an orientation mismatch at the UE, i.e., with an azimuth orientation of −30° and 30°, results in abrupt decreases in the RSRP measurements. Beam response characterization or iterative fitting can be used in the future to enhance the AoA performance of these reduced antenna arrays. Future work can also investigate the exploitation of the 5G mmWave testbed with multiple base stations.