.png){kind=icon}
Return
Accelerated Spectrogram Assembly and Processing
.png){kind=icon}
PRE
AI
DOI：10.1109/lsp.2026.3740634.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper presents Accelerated Spectrogram Assembly and Processing (ASAP), a method for low-latency spectrogram assembly from recursively synthesized spectral representations of time-series data. During acquisition, $N$-point discrete Fourier transforms (DFTs) are computed for fixed-length base blocks, and transforms for larger dyadically aligned superblocks are synthesized recursively and stored. Unlike conventional short-time Fourier transform (STFT) methods, which recompute DFTs for each analysis frame, ASAP reuses stored block transforms to assemble spectrogram columns. Phase-shifted stored block transforms represent conceptual time-domain intervals without explicitly constructing them. Segment placement and duration may be selected at analysis time on the base-block grid within configured bounds. Frame transforms are derived using either a rectangular window or a segment-length-adaptive five-tap cosine-series taper that is symmetric and zero at segment endpoints. Only preselected output bands and the additional frequency bins required by the taper are retained. In a sparse-band validation, the worst peak-scaled error relative to direct double-precision references was $9.84\times 10^{-6}$, and incremental processing was 323–384 times faster per column than direct tapering followed by a full $N$-point FFT.
Keywords:
Spectrogram assembly
short-time Fourier transform (STFT)
time-frequency analysis
dyadic block hierarchy
low-latency signal processing
Journal
I
IF:
3.9
Papers:
800
Citations:
0
Organization
Cited Papers
No cited papers available