Intelligent MIMO Systems

Learning-based channel estimation, compression, and adaptive feedback for efficient MIMO networks

Multiple-input multiple-output (MIMO) networks depend on accurate and timely channel knowledge. In practice, acquiring that knowledge introduces substantial overhead: channel estimates are compressed, transmitted over limited feedback links, and quickly become outdated as propagation conditions change. Measurements from a single frequency band may also hide important multipath components through frequency-selective destructive interference.

This research program develops intelligent channel-acquisition mechanisms that improve what channel information is measured, how efficiently it is represented, and when it should be transmitted. The work spans multi-band channel fusion, learning-driven feedback scheduling, and experimental analysis of standardized MIMO feedback compression.

BANDWEAVE: learning to fuse channels across bands

BANDWEAVE—recipient of the Best Paper Award at IEEE INFOCOM 2026—is a multi-band CFR fusion framework that reconstructs a more complete view of the wireless channel from complementary observations across frequency bands. Unlike spectrum-aggregation methods designed primarily for sensing or localization, BANDWEAVE directly optimizes channel estimates for end-to-end communication performance.

Its progressive learning pipeline has three phases:

  1. Supervised pretraining learns a shared multi-band channel representation.
  2. Simulation-in-the-loop fine-tuning connects the learned representation to physical-layer metrics such as bit error rate (BER).
  3. Online feedback-aware adaptation adjusts the fusion policy as propagation conditions change.

BANDWEAVE was evaluated on an IEEE 802.11ac MU-MIMO Wi-Fi testbed and a 60 GHz mmWave MIMO platform across three propagation environments. It delivered more than 16% improvement in throughput and BER, achieved over 4.9× communication-performance gain relative to the evaluated band-merging approaches, and reduced inference time and energy consumption by up to 17× and 18×, respectively, on resource-constrained edge platforms. The implementation is open source.

BANDWEAVE combines CFRs from multiple contiguous or non-contiguous bands, progressively learns a communication-aware fusion policy, and feeds the enhanced channel estimate back for MIMO precoding.

SHRINK: feedback only when the channel requires it

IEEE 802.11 channel sounding normally requests feedback at fixed intervals, even when the wireless channel is stable. SHRINK replaces this rigid behavior with a data-driven sounding policy. Each station analyzes its current and previous channel estimates, predicts the resulting throughput variation, and decides whether to transmit updated feedback or a lightweight negative acknowledgment.

Experiments with commercial Wi-Fi devices—including measurements in an anechoic chamber and dynamic indoor environments—showed that SHRINK:

  • Reduced feedback airtime and data overhead by 81% on average without degrading precoding performance.
  • Improved overhead reduction by 33.6% on average over the evaluated state-of-the-art approaches.
  • Increased throughput by 24.5% through more efficient use of channel airtime.

The SHRINK implementation is publicly available.

SHRINK predicts whether reusing the previous feedback will affect throughput and transmits a new compressed channel estimate only when necessary.

Quantization and grouping in standardized MIMO feedback

Before optimizing when feedback should be sent, it is essential to understand how standardized compression changes the channel representation itself. Our IEEE 802.11 feedback study systematically evaluates beamforming-angle quantization and OFDM subchannel grouping in IEEE 802.11ac/ax networks.

The evaluation combines channel measurements from commercial devices with standards-compliant emulation across multiple MIMO configurations, bandwidths, modulation and coding schemes, and propagation environments. It characterizes the tradeoff between feedback overhead, beamforming-matrix reconstruction error, and BER, providing an experimental benchmark for future compression and channel-acquisition strategies. The datasets and emulation framework are publicly available.

Experimental evaluation of beamforming-feedback reconstruction error and BER across IEEE 802.11ac/ax configurations and propagation environments.

Publications

  1. K. F. Haque, F. Meneghello, J. Ashdown, and F. Restuccia, “BANDWEAVE: Enhanced Channel Estimation in MIMO Networks with Multi-Band Fusion”, IEEE INFOCOM, 2026. Best Paper Award. Code.
  2. K. M. Rumman, F. Meneghello, K. F. Haque, F. Gringoli, and F. Restuccia, “SHRINK: Reducing MIMO Feedback Overhead in Wi-Fi with Dynamic Data-Driven Channel Sounding”, ACM MobiHoc, 2025. Code.
  3. F. Meneghello, K. F. Haque, and F. Restuccia, “Evaluating the Impact of Channel Feedback Quantization and Grouping in IEEE 802.11 MIMO Wi-Fi Networks”, IEEE Wireless Communications Letters, 2024. Code and datasets.