One of targeted thrusts in this project is Tb/sec aggregate access to mobile clients by incrementally aggregating clients to enable simultaneous transmission to multiple clients while controlling inter-stream interference. Transmissions in millimeter-wave spectrum need to be directional and thus both transmitter and receiver needs to discover suitable directional beams to enable a point-to-point directional link. In a downlink multi-user scenario, we have to identify clients for simultaneous transmission such that inter-stream interference due to spurious reflections is minimal. This project studies the general problem of user and beam selection for enabling multi-user simultaneous directional transmission with minimum inter-stream interference. The key activity was to design and experimentally evaluate user and beam selection mechanisms that can successfully enable multi-user downlink transmission given the propagation characteristics of millimeter-wave spectrum.
In order to maximize the aggregated capacity of a multi-stream transmission, analog beams need to be jointly configured with user selection and digital weights (driven from a interference cancellation method such as Zero-Forcing). In this project, we first formulate the capacity under joint user and beam selection; however, prohibitively large training and feedback overhead associated with this joint maximization makes it impractical for real implementations especially in mobile networks with large client population.
Second, we design a low-complexity structure for decoupling beam steering and user selection. This structure allows for finding analog beam steering parameters between the access point and each client, independent of potential grouping or multi-user transmission. Decoupling beam steering and user selection facilities scaling the network since the access point repeats the high-overhead beam training as required (when the beams are not reliable anymore due to mobility) while performing user selection for every multi-user transmission.
Third, this project targets to design user selection strategies to be utilized in the decoupled structure. To this end, we proposed two classes of user selection strategies; namely, single-shot and incremental. In single-shot class, users are grouped solely based on the information acquired in beam steering in one epoch without further channel sounding or feedback exchange. The rationale is that the sparse-scattering nature of spectrum makes it possible to mitigate inter-user interference solely via beam steering. One example of such user selection strategy is grouping users with lowest beam overlap by utilizing the reports from the prior beam steering.
Fig. above: Single-Shot (S²) User Selection in the context of analog and digital beamforming.
In incremental class, users are selected via a multi-round procedure in which each round includes AP acquisition of empirical interference information and the achievable sum-rate after zero-forcing for users included in the round. The AP tests and trains one or more users per round, and incorporates the net positive effect of adding a user with the determinant of additional inter-user interference. While incremental user selection can potentially achieve higher sum-rate due to additional multi-user interference measurements, it requires larger overhead compared to single-shot approaches.
Figure above: Interference-aware Incremental (I²) User Selection in the context of analog and digital beamforming.
We performed an extensive measurement and simulation study using steerable 60 GHZ RF-fronted combined with the software-defined radio platform WARP. First, we studied the maximum achievable throughout via joint user and beam selection by exhaustive search over all possible user and beam combinations. We showed that user aggregation can scale the network throughout in millimeter-wave via joint user and beam selection approach. Specifically, we demonstrated that the AP is able to successfully aggregate 12 spatial streams in different Line-of-Sight (LoS) and Non-Line-of-Sight (NLoS) scenarios.
Figure left: 60 GHz hardware blocks. Figure right: The 60 GHz setup in a conference room where the data was collected.
Second, we implemented decoupled beam steering and user selection and performed similar experiments in different environments in order to compare joint and decoupled structures. We found that despite its improved simplicity and lower complexity, decoupling beam steering and user selection incurs less than 5% capacity loss with four RF chains at the AP.
Third, we evaluated single-shot and incremental classes of user selection. We found that in two-user or three-user aggregation, the single-shot strategies attain comparable throughput with incremental strategies. However, as the number of aggregated clients increases, the excessive inter-user interference increases the performance gap between these two classes of user selection. Furthermore, we studied the NLOS scenarios where the LoS path from the AP to client is blocked. We found that the lower SNR of NLOS users makes them less tolerant to interference and thus single-shot user selection does not perform well for such scenarios.
Figure above: Achievable sum capacity with S² and I² policies under LOS and NLOS link connectivities with a four-RF chain AP.
Figure above: Normalized capacity of S² and I² policies under LOS and NLOS link connectivities with a four-RF chain AP.
Y. Ghasempour and E. Knightly. (2017). Decoupling Beam Steering and User Selection for Scaling Multi-User 60 GHz WLANs. Proceedings of the … ACM International Symposium on Mobile Ad Hoc Networking & Computing.