Multi-Stream Beam-Training for mmWave MIMO Networks

Multi-stream communication in the 60 GHz band has the potential to achieve up to 100 Gbps via concurrent transmission of multiple data streams. Unfortunately, establishing multi-stream directional links is a high overhead procedure in which the search space scales with the number of spatial streams and the product of AP-Client beam resolution. Hence, the key challenge to realizing such rates is to efficiently discover the analog beams at the Access Point (AP) and clients that support concurrent directional transmission/reception of multiple data streams without excessive inter-stream interference. The key activity over the past year was to design, implement and experimentally evaluate an efficient beam steering protocol for downlink multi-user MIMO transmission in 60 GHz WLANs.

The beam steering has a significant role in the achievable throughput of a multi-user 60 GHz transmission since (i) it provides SNR boost via beamforming gain to invoke higher order modulations; and (ii) it impacts the channel orthogonality between a pre-selected group of clients. The latter is the case due to sparse scattering of millimeter waves, i.e., only a few dominant LOS and NLOS paths characterize the channel between any two nodes. An analog beam acts as an amplifier, boosting the strength of certain paths within its main lobe (and side-lobes) and weakening the others. Hence, analog beam steering impacts the effective channel between the AP and each client, and consequently the achievable multiplexing gain.

Figure above: Aggregate PHY rate of a two-user MIMO transmission to R1 (fixed at position index 1) and R2 when placed at other 11 positions.

In millimeter-wave networks, the AP needs to establish and maintain a directional link with every client in the network. This is achieved by a beam acquisition procedure that includes periodic transmission of multiple training frames. In this project, we first reuse the received training frames to fully resolve the dominant LOS and NLOS paths between the AP and each client and their relative timing (i.e., power delay profile or PDP). This is particularly possible thanks to the high sampling rate at 60 GHz and sparse channels.

Unfortunately, solely adding PDP information does not solve the multi-stream beam selection problem as it does not contain any direction information. Second, we couple the knowledge of beam patterns in the AP’s RF codebook with PDP estimates for each beam to infer the direction of each dominant path. In other words, by weighting each PDP to the known directional gain for that beam pattern, we can narrow down the direction interval that each path may fall into.

Figure above: Experimental floorplan. Square boxes represent client positions.

Third, we leverage the collected direction inferences for all intended clients to select a set of analog beams with diverse or ideally orthogonal paths for downlink multi-stream concurrent transmission. The beams are selected such that they provide highest SNR at the target client while maximizing channel orthogonality between clients.

Significant Results:

We have implemented and evaluated our design using a programmable testbed for wideband 60 GHz WLANs with electronically-steerable phased arrays. We conducted over-the-air experiments in different indoor settings and our key findings are as follows:

The practical 60 GHz beams generated via phased array antennas have highly irregular beam patterns. Nonetheless, despite their irregularity, the directivity gain is known a priori in each direction as it is a deterministic function of the codebook and antenna spacing. Due to this irregularity, several beams may capture the same LOS/NLOS path through their main-lobes or side-lobes. Multiplexing independent data streams over such beams hinders the multiplexing gain as their channel is correlated due to the common dominant path and sparse nature of 60 GHz propagation. Applying digital precoding (e.g., zero-forcing) can mitigate the inter-stream interference in such cases; however, it cannot fully compensate for high channel correlation and low stream separability.

Figure above: The aggregate PHY rate as a function of the number of spatial streams.

We show that our design successfully obtains multiplexing gains for both single-user MIMO and multi-user MIMO in the evaluated scenarios. Our results demonstrate that the aggregate PHY throughout is in average within the 90% of maximum achievable throughput (being realized by an exhaustive search over all possible AP-clients beam combinations) in different multi-client settings with 2 to 4 concurrent transmissions. Yet, our system overhead is a linear function of number of multipath components in the sparse 60 GHz band; hence, it requires only 0.04% of the exhaustive search’s overhead.


Y. Ghasempour, M.K. Haider, C. Cordeiro, D. Koutsonikolas and E. Knightly, “Multi-Stream Beam-Training for mmWave MIMO Networks,” in Proceedings of ACM MobiCom 2018,New Delhi, India, October 2018.