Millimeter-Wave System with a Diverse Spectrum Architecture

Carrier frequencies at 60 GHz and above provide the necessary spectral width and propagation characteristics to scale the WLAN capacity to Tb/sec using WATabit nodes. Unfortunately, millimeter scale wavelengths have extremely poor ability to penetrate objects. For example, brick material attenuates a 60 GHz signal by 14 dB/cm yielding a near complete blockage for a 70 cm brick wall (980 dB) whereas at 500 MHz carrier frequency in the UHF band, a 70 cm brick wall only attenuates 10 to 15 dB total. The project’s vision is to overcome the propagation limits of millimeter wave by exploiting spectrum diversity that spans over two orders of magnitude (100x), from legacy unlicensed and UHF whitespace bands (500 MHz to 5 GHz) to millimeter wave (30 GHz to 300 GHz). The design will enable a device with a single legacy-band antenna to spoof legacy-band MIMO infrastructure into performing full-rank transmission and reception, thereby achieving a combination of rate, range and obstacle penetration that is otherwise impossible. The key technique is to form two coupled antenna arrays separated by 100x spectrum, utilizing the high rate and synchronization of WATabit node to share both data and Channel State Information (CSI) to harness other nearby legacy-band antennas thereby enabling a legacy-band virtual array.

The major activity over the past year was to design a mm-wave system with a diverse spectrum architecture that includes spectrum in the visible light band (430-770 THz range) along with the mm-wave band and to experimentally evaluate the system to achieve high data rate links despite device mobility (a key challenge limiting the throughput of mm-wave transmissions in practical systems). In particular, we develop a mm-wave system which tracks device mobility by passively sensing changes in light intensity from off-the-shelf light sensors to obtain direction estimates which is further used to infer changes and adapt the mm-wave beams at WATabit nodes in order to maintain high SNR links despite device mobility. An earlier version of our design where we adapt beams at both the WAtabit node and client ends, requires feedback packets for beam adaptation and localization at WATabit and maintaining LOS with multiple ( 3) light sources presents greater challenge in terms of infrastructure requirements and LOS blockage. In contrast, our current design exploits a single light source collocated at the WATabit node to adapt client beams without any feedback and handles beam adaptation by repurposing periodic beacon-sweeps. Furthermore, we have also enhanced the protocol design to include multi-client beam steering to address beam adaptation at WATabit node, extended the hardware implementation to include not only horn antennas but also the practical phased arrays with irregular beam patterns which impact the optimality of LOS path based geometric beam steering, and evaluated in WLANs with multiple clients (including impact of contention) and various indoor mobility patterns. We demonstrate that this system maintains beam alignment at the narrowest beamwidth level even in case of device mobility, without incurring any training overhead at mobile devices. As such, this indicates that our system has robust and extremely fast mm-wave links to share data and CSI with very low latency, necessary to form a virtual array in the legacy band and thereby achieve uplink spatial multiplexing.

The key requirements to support full rank spatial multiplexing via virtual array of single-antenna legacy devices is that WATabit nodes should be able to maintain sufficient Line of Sight (LOS) stability in mm-wave bands in order to achieve the real-time data transmission and reception. However, the high directional nature of mm-wave links achieved via horn antennas or large phased array antennas to battle the higher path-loss makes the mm-wave communication sensitive to link misalignment due to nodal mobility such as rotation and translation. To restore and maintain alignment, nodes may require to undergo frequent beam search by exhaustively sweeping across beams available at both ends to re-establish a high data rate link. Addressing the frequent misalignment therefore results in beam search overhead which has a detrimental effect on the mm-wave link throughput. This mobility induced link breakage is a key challenge in our system design as it can interrupt the communication in mm-wave bands, consequently affecting the rank of spatial multiplexing in the legacy bands.

To maintain LOS stability even in presence of device mobility, we devise a diverse system architecture that comprises two distinct bands; a Communication Band and a Sensing Band. The former comprises of mm-wave band radios and mm-wave antennas at the WATabit nodes and client nodes, over which data communication takes place. The Sensing band (430-770 THz frequency range) exploits indoor light sources as fixed anchors, and track changes in mobile devices’ position and orientation by passively sensing light intensity from these indoor luminaries using off-the-shelf light sensors. We then translate these position and orientation estimates to changes in LOS path parameters in the mm-wave band and continuously infer required changes in mm-wave antenna beams without requiring any beam training at the client devices. The design also supports a new feature such that the WATabit node can simultaneously align beams with multiple clients by performing a beam sweep only once at its end, with client beams selected via out-of-band light sensing. In particular, we complete the following three objectives in mm-wave band part of overall system design.

First, we devise a novel method for incoherent-light Angle of Arrival (AoA) estimation by using an array of light sensors. Our design is motivated by two key observations: First, most off-the-shelf wireless APs are equipped with light sources such as notification LEDs, which are in close proximity to their mmWave antennas. Second, mmWave channels exhibit pseudo-optical properties due to very short wavelength, i.e., dominant Line of Sight (LOS) propagation, limited scattering and reduced multipath. Therefore, our key idea is that by estimating the AoA corresponding to the LOS path from the AP’s indicator LED using light intensity measurements, we can approximate it as the AoA in the mm-wave band due to the close proximity of the AP’s LED and mmWave band antenna. This method doesn’t require any calibration or knowledge of the AP’s position or client’s orientation. Moreover, our method estimates AoA in both the azimuth and elevation planes, allowing us to steer beams for both 2-D and 3-D beamforming codebooks.

Second, we design an algorithm to steer mm-wave beams at WATabit nodes to acquire and maintain beam alignment at the narrowest beamwidth level, enabling highest data rates and consequently full rank spatial multiplexing in the legacy bands despite device mobility and without requiring any beam training at the client devices. The key idea is to exploit the aforementioned WATabit node’s light-based mobility estimates of LOS path at the client and then select the client-side beam with maximum directivity gain along the estimated AOA. As such, we avoid any mmWave in-band training or beam-search at the client in the presence of LOS path. In particular, the beam alignment protocol has two phases i) Beam Acquisition phase in which a client estimates its maximal strength beam using the light measurements as described in first step and uses this beam to receive in mm-wave band while the WATabit node does a beam sweep at its end. The client then sends feedback about the maximal strength beam to WATabit node. ii) Beam Steering phase in which the WATabit node passively tracks the AoA and continuously estimates and adopts the best client-side beam using the AoA estimates of the LOS channel. In addition, our design is scalable, such that the WATabit node can simultaneously align beams with multiple clients by performing a beam sweep only once at its end, with client beams selected via out-of-band light sensing. Moreover, this mobility tracking and beam steering is completely local to WATabit nodes, such that the legacy-band AP is oblivious to any changes, in accordance with the overall design principle of spoofing a virtual array using multiple WATabit nodes in the legacy bands.

Third, we implement our design on a custom dual-band hardware testbed and perform extensive over-the-air experiments in various environments and under different mobility scenarios to evaluate key components of our design. Our hardware platforms encompass off-the-shelf light sensors for light sensing, horn antenna based transceivers with 7o beamwidth to achieve extremely directional links, and X60, a 60 GHz wideband platform with an electronically steerable phased array for evaluation with practical antenna arrays which exhibit non-uniform beam patterns and side-lobes as they have an impact on the optimality of LOS path based geometric beam steering. We also develop a trace- and model- driven WLAN simulator to explore a broader set of operational conditions beyond the capabilities of our hardware platform, including multiple clients, different mobility patterns and client speeds.

We implement our design on a custom dual-band hardware platform comprising both mm-wave band transceivers with highly directional horn antennas as well as electronic beam steering enabled practical phased array antennas and indoor light sources and perform extensive over-the-air experiments in various environments and under different mobility scenarios. We first demonstrate that our incoherent-light intensity based position and orientation method tracks device mobility with high accuracy in practical indoor environments and mobility scenarios. In particular, our method estimates AoA within 3.5o of the ground truth for more than 90% of measurement instances in over-the-air experiments, demonstrating the viability and high accuracy of our estimation method. Further, higher accuracy can be realized by placing sensors close to device edges to further reduce inter-sensor distance, a key factor affecting the estimation accuracy.

Second, we study the beam steering capability of our system design by using the mobility estimates computed using visible light measurements. For this, we setup the dual-band testbed in such a way that the WATabit node is fixed, while we consider multiple positions and orientations for the client to study both translational and rotational mobility. Furthermore, we repeat the same experiment with both horn antenna and phased array platforms to compare the performance of different mm-wave band systems. Our results show that with highly directional beams of 7o wide horn antennas, WATabit node acquires client beams to within 1 sector of the true highest strength sector without any in-band training. Further, by estimating changes in AoA from the light source, we are able to track rotation leading to almost 97% steering accuracy when AoA estimates are computed at a modest rate of every 1o of client rotation. On the other hand, even with phased antenna arrays with nonuniform beam patterns and strong side-lobes, we are able to acquire client-side beams within 2 indices of the optimal beams for more than 80% instances, with AoA estimates almost always selecting the same beam as the true geometric AoA. Moreover, the SNR loss is within 1.5 dB compared to an exhaustive search over all beams, which incurs significant overhead and interrupts data communication, showing that there is only a marginal gain for performing exhaustive search at the client-end. Without beam adaptation, mm-wave links can lose multi-Gbps data rates via a mere 0.5m translation despite wider beam patterns of a practical phased array, highlighting their susceptibility to client mobility. With our light assisted beam steering, our phased array antenna platform experiments with client translation and rotation further show that the client maintains a highly directional link with SNR within 1 dB of exhaustive search for most positions along the trajectory via AoA estimation.

Finally, to explore a broader set of operational conditions beyond the capabilities of the hardware platform including WLAN performance with multiple clients, multi-client beam training to address AP-side beam adaptation and different mobility patterns and speeds, we also develop a custom WLAN simulator. We use channel traces from our hardware experiments to drive the simulator and evaluated in WLANs with multiple clients (including impact of contention) and various indoor mobility patterns. Our extensive experiments demonstrate that our design incurs negligible overhead for polling after the mandatory beacon sweeps by the WATabit node, with client-side beams acquired and steered using AoA estimates only. Thus, its overhead also remains nearly constant despite an increase in the number of clients, incurring 10x to 15x lower overhead than an 802.11ad based baseline scheme with in-band retraining. Furthermore, due to repeated beam training overhead in 802.11ad, more than 50% of available throughput is lost due to beam misalignment from translational and rotational mobility. By light assisted beam steering, we demonstrate up to 2x gain in throughput by avoiding training overhead in client-side adaptation in majority of the cases and moreover, its performance scales much better with rotational speed and the number of client devices.