.png){kind=icon}
.png){kind=icon}
Return
Key technologies for ultra-high throughput neural signal recording: A graphical data compression method based on two-dimensional compressed sensing
.png){kind=icon}
PRE
AI
DOI：10.1016/j.bspc.2024.107448.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
As the number of channels in wireless intracortical neural recording systems continues to increase, the bandwidth and power consumption constraints of the wireless data link become more stringent, necessitating ondevice data compression. However, effectively reducing the wireless payload while preserving the maximum amount of information remains a challenging problem that has yet to be fully addressed. This paper proposes a graphical data compression scheme based on two-dimensional compressed sensing (2D-CS) designed for ultrahigh channel neural recording systems. The scheme comprises two main components: a unified matrix representation of neural signals and 2D-CS. Unlike previous schemes that focus solely on compressing repetitive single-channel or localized multi-channel neural signals, this approach converts high-throughput one-dimensional neural data into a unified graphical representation using a matrix-based format. This method leverages the rich inter-channel correlations afforded by the increased number of channels and applies 2D-CS to compress the graphical data in its entirety, thereby substantially improving compression efficiency. The efficacy of this method was validated using real datasets from a 1024-channel system, encompassing both action potentials (APs) and local field potentials (LFPs). The results indicate that after compression with this method, all APs remained intact, the percent root mean square deviation (PRD) of the LFPs was limited to 5.52 %, the average signal-tonoise and distortion ratio (SDNR) was 29.5 dB, and the overall compression ratio (CR) achieved was 84. This outcome provides a reliable data compression solution for implantable brain-computer interfaces (iBCIs) accommodating large channel counts and supports the development of low-power devices.
Keywords:
Wireless iBCIs
Neural signal processing
LFPs
Data compression
2D-CS
APs
Journal
IF:
4.9
Papers:
9.8K
Citations:
2.4W