Functional magnetic resonance imaging (fMRI) provides
high-resolution, voxel-wise measurements of brain activity, but
acquiring large-scale fMRI datasets is expensive, immobile,
and time-consuming. At the same time, commodity wireless
devices continually capture channel state information (CSI)
— a rich, multi-dimensional signal that reflects environmental
changes and human motion. This paper investigates whether
carefully processed CSI can serve as a low-cost, portable
proxy for coarse voxel-wise neural activation in controlled
experimental paradigms.
Our goal is not to replace fMRI, but to explore whether CSI
contains enough information to reconstruct low-dimensional
summaries of neural activity and to detect block-like activation
patterns under controlled conditions. If successful, such a
proxy could enable inexpensive, mobile monitoring and rapid
prototyping of neuroimaging-inspired interfaces.
Challenges. Mapping CSI to voxel-like signals presents
several technical challenges: (i) differing sampling rates and
clocks (Wi-Fi CSI is sampled at high frequency while fMRI
TRs are much slower), (ii) unknown and time-varying delays
(clock offsets and drift) between sensors, (iii) heterogeneity
across subcarriers and receiving antennas, and (iv) severe,
structured noise due to multipath and non-neural motion.
Addressing these requires robust preprocessing, alignment,
and decoder design that tolerate misalignment and domain
mismatch.
Approach. We present a simulation-led pipeline that synthesizes paired CSI and voxel-like time series under controlled
offsets and drift, applies alignment and time-warping to synchronize signals, and trains simple linear decoders to predict
voxel activity from aggregated CSI features. The pipeline
produces three primary figures: (1) alignment timelines (before/after), (2) per-voxel correlation distributions, and (3) ROC
curves for block-activity detection. All scripts and synthetic
data are provided so results are reproducible and the pipeline
can be reused as a test harness.
Contributions. This work makes four concrete contributions:
- A compact, reproducible simulation and processing
pipeline that generates paired CSI and voxel signals with
configurable offsets and drift. - A lightweight alignment method (lag estimation + linear
time-warp) that corrects clock offsets and drift between
modalities. - An empirical evaluation showing that aggregated CSI
features, coupled with simple ridge decoders, recover
coarse voxel activity and detect block activations with
non-trivial AUC in simulation. - A small, self-contained LaTeX project (scripts, figures,
and captions) that demonstrates the pipeline and provides
a press-style build target for rapid iteration.
Outline. The remainder of the paper is organized as follows.
Section II describes the synthetic data generation, feature extraction, and alignment procedures. Section III details the experimental settings and evaluation metrics. Section ?? presents
the alignment, correlation, and ROC figures, and Section ??
discusses limitations and next steps toward real-world CSI-tovoxel evaluation.
Updates In Progress: