Flash-Attention MHLA for RF SpectrumCompression: SpectrumEncoder with Token-Dropout and RoPEAblations

We present a lightweight SpectrumEncoder forcompressing FFT power spectra using multi-head linear attention(MHLA) with FlashAttention backends and token-dropout. Wereport compression–accuracy trade-offs, latency profiles, and anablation on Rotary Positional Embeddings (RoPE). The method isdesigned for real-time SIGINT pipelines where millisecond-levellatency and energy budgets matter, enabling up to 40% moreconcurrent RF bands on the same hardware.