One of the targeted thrusts in this project, Thrust 2 is to address the challenge of uplink multi-user stream separation in S-T WLANs lacking rich multi-path. While clients may be spatially separated, our initial experiments at 60 GHz indicate that their simultaneous transmission to a single point in space (the Access Point (AP)) will not be feasible, even with receive beamforming. Hence, we will explore the multi-beam codebooks as a mechanism to provide sufficient diversity to enable the AP to separate streams and manage inter-stream interference, also exploiting the limited reflections available in NLOS paths.
Due to increasing disparities in the capabilities and form factors of APs vs. clients, it is increasingly clear that data rates must be scaled not solely via a high point-to-point PHY rate, but also via simultaneous transmission and reception to and from multiple clients. While today’s millimeter wave (mmWave) WLANs can realize simultaneous multi-user multi-stream transmission mainly by exploiting directional transmission and physical separation of clients, these systems are solely limited to downlink. In contrast, simultaneous mmWave multi-user/ multi-stream uplink transmission must address the inevitable interference from clients directing their transmissions towards a common point in space, namely, the receiving AP. In order to support simultaneous multi-user multi-stream reception, we design and experimentally evaluate Uplink Multi-User Beamforming via a Single RF chain AP, the first system for multi-user mmWave uplink.
In order to support uplink multi-stream multi-user transmissions, we make the following contributions.
First, we propose a 60 GHz WLAN architecture for a multiuser uplink using only a single RF chain at the AP. That is, the AP has a phased array for receive and transmit beamforming but does not have MIMO. Transmission is initiated by the AP with a downlink trigger frame as employed by standards such as IEEE 802.11ax. After the trigger frame, we stagger uplink client PHY preambles so that the AP can obtain a “clean” (interference-free) channel measurement for each user to be used during decoding. Subsequently, the triggered clients transmit their uplink data frames in parallel. While these frames are temporally aligned by the trigger, they arrive at the AP offset by the clients’ different propagation delays. Consequently, we design our system to enable asynchronous decoding, i.e., we do not require symbol-level synchronization. To realize this feature, we design Scalable Multi-User Overlayed Constellations. In particular, we overlay Amplitude and Phase Shift Keying (APSK) constellations such that each user is assigned one or more consecutive rings and groups of rings are assigned to users such that the highest SNR user has the outermost ring. With sufficient SNR spread among the rings, the AP can then successively decode one user at a time starting with the highest SNR user, i.e., we enable the use of Successive Interference Cancellation (SIC) decoding. We show that with this multi-user overlay strategy, at each stage of stream separation, the current symbol being decoded on a particular nearly constant amplitude constellation ring is resilient to the detrimental impact of phase noise impairment caused by interference from other streams which are being received at significantly different amplitudes. Moreover, we design a Carrier Frequency Offset (CFO) compensation method comprised of pre-compensation and iterative correction. This allows the AP to apply the offset of each user to the composite stream at each interference cancellation iteration, while treating the rest of the signals as noise. When decoding the signal from one user, the AP employs an interference alleviation filter specifically designed from the training preamble of that user to cancel the interference and recover the signal.
Second, we show how to use beam selection to attain the desired ring separation and hence, SNR separation, at the access point in order to realize high aggregate rate. We show how both AP and client beams can be re-steered to maximize the aggregate multi-user rate using the outcome of single user training, i.e., without transmission of additional training frames. Since our system is constrained with to receive data on a single RF chain, the beam selection is constrained in multiple ways: (i) the AP must use a single receive beam from its codebook to receive the superposition of the data streams without MIMO processing due to having only a single RF chain. Nonetheless, both the AP and all clients can steer their beams in directions that yields maximum aggregate rate. Unfortunately, (ii) the search space for considering all such combinations may render joint optimization impractical, as it in principle requires exhaustive testing of all AP-user beam combinations. Therefore, we present three complementary beam selection policies with different computational requirements, spanning from testing all AP and user beam combinations, to only letting the AP re-steer its beam.
Finally, we implement the key components of our system using X60, a programmable testbed for wide-band 60 GHz WLANs with electronically steerable phased arrays. Moreover, we also deploy a WARP-60 testbed using a steerable 60 GHz RF-frontend combined with the software defined radio platform WARP. This platform utilizes mechanically steerable horn antenna with configurable beamwidths. Using these two testbeds, we perform over 67,000 over-the-air measurements and subsequently perform trace-driven emulations to study the multi-user performance of our system.
We performed extensive measurements using the X60 testbed and the WARP-60 testbed and presented the first experimental evaluation of uplink multi-user multi-stream transmission on a single RF chain in mmWave networks. We implemented the key components of our system and performed experiments in different indoor environments.
The spread in signal strengths among concurrently transmitting users affects the successful decoding of the composite stream at the AP. More specifically, the relative signal strength of all users as determined by the beam selection can be influenced by geometry of users and this affects the SIC user decoding order by allowing the AP to first decode the stronger and more robust users, therefore, reducing the number of errors propagated from one stage to the other. To demonstrate this, we consider a simplified setting of two users transmitting in the uplink and study the impact of geometry in terms of distance and angular separation of the users on the ability of the AP to decode the different user streams. We conduct over-the-air experiments using the X-60 testbed and experimentally explore the multi-user gains of our system in comparison to Single User transmission scheme.
Our experiments demonstrate that with beam resteering at the AP and at least one of the grouped users, our system yields aggregate rate gains of up to 1.45 x over Single User irrespective of the choice of the user group and the geometric separation between them.
Second, we study the critical role of SNR spread among concurrently transmitting users as determined by multi-user beam selection and show how it helps in limiting inter-user interference and leads to increased SINR for each user and increased gains of our system.
Third, we vary the receive bandwidth from wide to narrow and find that adapting beamwidth at the AP acts as a knob in controlling the SNR spread and grouping efficiency and thereby the aggregate rate for our system. We explore the tradeoff that wider beams, which can create Non Line-of-Sight (NLOS) paths even when a LOS path exists, can provide better channel grouping opportunities, as more users will be able to share a beam. To explore this beamwidth signal coverage tradeoffs, we employ the WARP-60 testbed which generates directional beams of varying beamwidth from 7o to 80o using horn antennas. We showed that increasing receive beamwidth at the AP leads to larger SNR spread among the users thus providing more beam sharing possibilities and hence increase aggregate rate for multi-user uplink transmission. Thus, despite its low computational overhead, we demonstrated that our method using 80o beamwidth achieves more than 1.5x multi-user capacity gains over single user transmission scheme.
Publications
K. Dasala, J. Jornet, and E. Knightly, “Uplink Multi-User Beamforming on Single RF Chain mmWave WLANs,” in Proceedings of IEEE INFOCOM 2021, May 2021. Best Paper Award.