Compressed Beamforming Matrix
The Compressed Beamforming Matrix is the mathematical structure used to encode Channel State Information (CSI) in a compact, transmittable form during the Wi-Fi beamforming feedback process. Rather than transmitting a full complex-valued matrix describing the wireless channel, a client device compresses the relevant spatial information into a small set of rotation angles and returns this as part of the Beamforming Feedback Information (BFI) frame. This compression is fundamental to the operation of IEEE 802.11 ac and ax (Wi-Fi 5 and Wi-Fi 6) multi-antenna systems — and, significantly, it retains enough environmental information to enable passive sensing of human presence, movement, and activity, even when intercepted by third parties.
Definition
The compressed beamforming matrix is a compact representation of the matrix V — specifically, the right singular vectors of the wireless channel matrix H. These singular vectors encode the preferred spatial directions along which transmitted energy will most efficiently reach the receiving client.
Rather than sending the raw complex numbers of V across the air, the client device encodes V as a sequence of Givens rotation angles: two angle parameters, conventionally labelled φ (phi) and ψ (psi), per rotation. These quantised angles are what actually appear in the Beamforming Feedback Information payload.
The result is a drastic reduction in feedback bandwidth. A full complex matrix for a 4×4 MIMO system at 256 subcarriers would require thousands of floating-point values; the compressed form reduces this to a handful of quantised integers per subcarrier group.
Mathematical Background
The Channel Matrix and SVD
The wireless MIMO channel between a transmitting access point (AP) and a receiving client is described by a matrix H, whose dimensions reflect the number of transmit and receive antennas. At any given sounding subcarrier, H is a complex-valued matrix.
Channel estimation allows the client to measure H from a received Null Data Packet (NDP). The client then performs a Singular Value Decomposition (SVD):
H = U Σ V^H
Where:
- U is a unitary matrix of left singular vectors (receive-side spatial directions)
- Σ is a diagonal matrix of singular values (signal strengths along each spatial direction)
- V^H is the conjugate transpose of V, a unitary matrix of right singular vectors (transmit-side spatial directions)
The columns of V represent the optimal transmit directions — the spatial modes along which the AP should steer energy to maximise throughput or signal quality at the client. The AP needs V to compute its transmit beamforming steering matrix, but it cannot measure V directly from its own end. The client must therefore feed this information back.
Why Only V?
The AP already knows its own antenna geometry. What it lacks is knowledge of the channel as seen from the client's perspective. The matrix V encapsulates precisely the transmit-side geometry needed to steer beams correctly. Transmitting U or Σ is unnecessary for this purpose — only V (or an approximation of it) is required.
Compression Mechanism
Givens Rotations
A unitary matrix like V can be decomposed into a product of Givens rotation matrices — elementary 2×2 rotations that operate on pairs of dimensions. Each Givens rotation is parameterised by two real angles:
- φ (phi): a rotation angle
- ψ (psi): a phase angle
By chaining together a specific sequence of these rotations (determined by the matrix dimensions), any unitary matrix can be reconstructed exactly. The client computes the Givens decomposition of V and extracts the corresponding angle pairs.
Quantisation
The continuous-valued angles are then quantised to a fixed number of bits before transmission:
- φ angles are typically quantised to b_phi bits
- ψ angles are quantised to b_psi bits
The IEEE 802.11 standards define specific quantisation precisions depending on the requested feedback quality level. Lower bit depths mean smaller feedback frames but coarser approximations of V.
Encoding
The quantised angle indices are packed into the Beamforming Feedback Information frame. The AP, upon receipt, uses these integers to reconstruct an approximate V by reversing the Givens rotation process.
Subcarrier Grouping
A full Wi-Fi channel may contain hundreds of OFDM subcarriers, each with its own channel matrix H and therefore its own V. Transmitting feedback for every subcarrier individually would be costly in airtime.
To reduce overhead, the standard allows subcarrier grouping via a parameter called N_g (grouping factor). Angles are reported once per group of N_g subcarriers rather than per individual subcarrier:
- N_g = 1: Full resolution — one feedback set per subcarrier (highest precision, largest frame)
- N_g = 2: One feedback set per pair of subcarriers
- N_g = 4: One feedback set per four subcarriers (lowest precision, smallest frame)
The grouping parameter represents an explicit trade-off: lower N_g values preserve fine-grained Channel State Information across the frequency band, while higher values sacrifice spatial resolution in exchange for reduced feedback overhead.
From a sensing perspective, even high grouping factors (N_g = 4) retain sufficient inter-subcarrier variation to support coarse environmental inference, though fine-grained gesture recognition benefits from lower grouping.
Decompression at the Access Point
When the AP receives the Beamforming Feedback Information frame, it performs the following steps:
- Dequantise the received angle indices back into floating-point angle values
- Reconstruct V̂ (an approximation of V) by multiplying the corresponding Givens rotation matrices in the specified order
- Apply V̂ as the precoding (steering) matrix for subsequent downlink transmissions to this client
The reconstructed V̂ is an approximation because quantisation introduces rounding error. The fidelity of this approximation depends on the bit depth of the quantisation and the grouping factor used. Higher precision feedback (more bits, lower N_g) yields a V̂ closer to the true V, improving beamforming gain.
The AP does not store raw channel matrices — it only receives and acts upon the compressed angle representation. This means the compressed beamforming matrix is the only form of Channel State Information the AP ever obtains from the client.
Surveillance Relevance
Environmental Information Preserved
Although the compressed beamforming matrix is a mathematical abstraction of the wireless channel rather than a raw recording of it, it preserves substantial environmental information. The singular vectors of H reflect how objects in the environment — including human bodies — scatter, absorb, and reflect Wi-Fi signals.
Researchers in Wi-Fi Sensing have demonstrated that:
- Human presence in a room produces characteristic changes in the dominant singular vectors of H
- Body movement shifts the angle parameters φ and ψ in ways that can be tracked over time
- Even with aggressive quantisation (e.g., 5–6 bits per angle), enough variation persists to classify activities such as walking, sitting, falling, and breathing
This means the compressed beamforming matrix — despite being designed purely for beamforming efficiency — functions as a compact but exploitable environmental fingerprint. Its smaller size compared to raw CSI actually makes it more practical to log and analyse at scale.
Sensing Without Raw CSI
Traditional Wi-Fi Sensing research often assumes access to raw complex Channel State Information as measured at the receiver. The compressed beamforming matrix opens a different attack surface: it is transmitted openly in the management frame exchange, requires no special hardware to receive, and can be collected passively by any device in range. This is a significant difference from active sensing approaches that require controlled transmissions or specialised receivers.
Interception
Unencrypted Transmission
Beamforming Feedback Information frames, which contain the compressed beamforming matrix, are management frames under the IEEE 802.11 protocol architecture. Like most 802.11 management frames, they are transmitted without encryption — even when the data channel between client and AP is protected by WPA2 or WPA3.
This means that any Wi-Fi adapter placed in monitor mode within radio range of the Channel Sounding exchange can capture the complete BFI frame, including the full compressed beamforming matrix.
Processing Intercepted Matrices
A passive interceptor receiving BFI frames can:
- Parse the frame to extract the quantised angle values for each subcarrier group
- Reconstruct an approximate V̂ using the same Givens decompression procedure used by the AP
- Apply Wi-Fi Sensing algorithms to the reconstructed matrix or directly to the angle time series
- Infer environmental state — presence, location, or activity — without the knowledge of the communicating parties
Because no decryption is required and no active transmission is necessary, this form of passive environmental sensing is essentially undetectable by conventional network security monitoring.
Implications for Through-Wall Surveillance
Because Wi-Fi signals penetrate walls, Channel Sounding exchanges between an AP and a client on one side of a wall produce beamforming matrices that are influenced by humans on both sides. An interceptor positioned outside a building can collect BFI frames from interior devices and reconstruct spatial information about the interior environment — a capability relevant to Through-Wall Surveillance research.
Standards
The compressed beamforming matrix format is defined in:
IEEE 802.11ac (VHT Beamforming)
- Supports up to 8 transmit antennas and 4 spatial streams
- Defines specific angle quantisation bit depths for different stream counts
- Maximum matrix dimensions: 8×4 (transmit × streams)
IEEE 802.11ax (HE Beamforming)
- Extends VHT beamforming feedback to OFDMA contexts
- Supports per-RU (Resource Unit) feedback, enabling finer-grained spatial feedback for multi-user scenarios
- Introduces additional feedback compression options for the larger subcarrier counts of 80 MHz and 160 MHz channels
Both standards are coordinated through the IEEE 802.11 working group. The compressed beamforming matrix format has remained structurally consistent across these generations, with incremental expansions in matrix dimensions and quantisation options.
See Also
- Beamforming Feedback Information
- Channel State Information
- Channel Sounding
- Channel Estimation
- Null Data Packet
- Through-Wall Surveillance
- Wi-Fi Sensing
- IEEE 802.11
- Surveillance Technology
References
- IEEE Std 802.11ac-2013 — Amendment 4: Enhancements for Very High Throughput for Operation in Bands below 6 GHz
- IEEE Std 802.11ax-2021 — High-Efficiency WLAN
- Tan, B. et al. (2021). "WiFi-based Passive Human Detection with Beamforming Feedback Information." IEEE Transactions on Mobile Computing
- Kotaru, M. et al. (2015). "SpotFi: Decimeter Level Localization Using WiFi." ACM SIGCOMM 2015
- Ma, Y. et al. (2019). "WiFi Sensing with Channel State Information: A Survey." ACM Computing Surveys
