Introduction
Wi-Fi sensing has emerged as a powerful technique for non-intrusive environmental monitoring, leveraging Channel State Information (CSI) to detect human gestures, respiration, and movement. Despite its potential, the field has been constrained by three significant "barriers": the "black-box" nature of commercial Wi-Fi hardware, pervasive hardware-induced CSI distortion, and the lack of flexible, high-performance software tools. This paper demystifies the internal baseband design of common Wi-Fi chips and introduces PicoScenes, a multi-threaded open-source platform designed to unlock the full potential of Wi-Fi sensing across both Network Interface Cards (NICs) and Software Defined Radios (SDRs).
Demystifying the Wi-Fi Baseband
To understand why CSI measurements often deviate from theoretical expectations, researchers must look inside the hardware. By reverse-engineering the Qualcomm AR9300 (QCA9300) architecture through driver analysis and hardware documentation, the authors mapped the complex three-stage transceiver path.
As shown in the architecture above, the signal passes through several critical stages, including the Analog Passband, Analog Baseband (I and Q paths), and Digital Baseband. A key discovery was the role of the BIQUAD filters within the analog baseband. These filters, intended for anti-aliasing and signal reconstruction, are the primary source of non-flatness in the frequency response.
The "Ghost" in the Machine: CSI Distortion
Researchers often assume that Wi-Fi hardware provides a "flat" view of the wireless channel. However, experiments in controlled environments (anechoic chambers and coaxial setups) reveal that internal filters introduce a predictable but significant distortion. This hardware-induced signature can be easily mistaken for environmental factors, leading to errors in sensing algorithms.
The distortion typically manifests as an "M-shaped" magnitude response and an "S-shaped" phase response. Crucially, this distortion is often asymmetrical, explaining why simple symmetric correction models frequently fail. By profiling these hardware-specific responses, the authors developed a distortion removal method to "clean" the CSI signal before it is used for sensing.
The PicoScenes Platform
To address the limitations of existing software, the authors developed PicoScenes, a unified middleware platform written in C++20. PicoScenes abstracts various hardware frontends—including the Intel IWL5300, QCA9300, and various SDRs—into a single, high-performance API.
Key innovations of the platform include:
- Arbitrary Frequency Tuning: By manipulating the baseband PLL and frequency synthesizer registers, PicoScenes can tune standard NICs to non-standard frequencies and bandwidths, effectively turning inexpensive Wi-Fi cards into wideband scanners.
- High-Performance SDR Baseband: For SDR users, the platform includes a custom-built 802.11a/g/n/ac/ax software baseband, allowing for full control over the physical layer (PHY) and supporting up to 200 MHz bandwidth.
- Rich Metadata: Beyond raw CSI, the platform extracts previously inaccessible PHY-layer metadata such as Carrier Frequency Offset (CFO), Sampling Frequency Offset (SFO), and Error Vector Magnitude (EVM).
Key Results and Impact
PicoScenes significantly pushes the boundaries of Wi-Fi sensing performance. The platform achieves a staggering 40 kHz CSI measurement rate on the QCA9300 and 1 kHz on SDRs. It has successfully unlocked over 2 GHz of available spectrum on standard hardware, far exceeding the limitations of standard Wi-Fi channels.
While the current SDR implementation is single-threaded, future work aims to parallelize the decoding flow to achieve even higher real-time throughput. By demystifying the hardware and providing a robust, open-source toolset, this work provides a new foundation for the next generation of high-precision wireless sensing research.
Summary updated