Wi-Fi MIMO

Multiple-Input Multiple Output (MIMO) is an essential part of any high-datarate wireless technology. Devices with multiple antennas can enjoy spatial diversity, which allows creating several spatial streams. They can be used for parallel data transmission (spatial multiplexing), focusing signal in the strongest stream (beamforming), and servicing multiple stations simultaneously (MU-MIMO). MIMO has a strong impact on Wi-Fi technology, and Wireless Networks Lab works in this direction from multiple perspectives.

Wireless Networks Lab has developed a flexible testbed based on multiple USRP software-defined radios. These devices are actively used across multiple research projects for conducting high-resolution channel measurements in indoor and dynamic environments, as well as for prototyping and over-the-air validation of novel channel state information (CSI) compression and sounding overhead reduction techniques.

MIMO can indeed provide large performance gains, provided that the transmitter has an accurate CSI. As the channel varies with time, the devices need to update CSI from time to time. However, the wider the channel and the higher the number of antennas, the heavier is the CSI feedback. We have developed a low-complexity lossless feedback compression method for 802.11be MU-MIMO networks. It is based on the standard representation of CSI and exploits correlations in both time and frequency domains. In a practical experiment in a dynamic indoor scenario, our method reduced the CSI size by 50%.

A complementary way to reduce the overhead from the channel sounding is increasing its period. Selecting an appropriate sounding period that balances the CSI freshness and overhead volume is a non-trivial problem. We were the first to optimize the explicit channel sounding period under dynamic MU-MIMO configuration. Through extensive simulation in ns-3, we identified subtle cases where the quasi-optimal period notably degrades throughput and provided recommendations on selecting the sounding period.

Implicit sounding is an alternative method for overhead reduction. Unlike in the standard explicit sounding, the client stations send a training signal instead of the AP, and the AP directly measures the reciprocal channel and obtains the CSI. Using the ns-3 simulator, we carefully compared explicit and implicit sounding mechanisms considering the wireless channel, the protocol overhead of transmitting channel information, and its distortion during measurement. We revealed that the gain of using implicit sounding is largely affected by residual non-reciprocity of channel measurements. To address this issue, we developed a calibration algorithm that compensates for this non-reciprocity using occasional bi-directional channel measurements.

Simply scaling hardware does not linearly improve performance in Wi-Fi networks, for example when servicing Virtual Reality applications. We extended the ns-3 simulation framework to support realistic VR traffic and up to 16 antennas at access points, confirming the diminishing returns of MIMO scaling. We also demonstrated that in highly dynamic environments performance is primarily limited by the equalizers at the stations, making SU-MIMO a surprisingly viable alternative. Most importantly, we implemented a delay-aware scheduler that increased the number of satisfied VR users by up to 50%, in some scenarios exceeding the benefit of doubling the antenna count.

List of relevant papers: