Skip to content

Imaging, DSP & Coding

Bitstream arithmetic shines when the output quality metric is perceptual (PSNR / SSIM) or algorithmic (BER after iterations), not bit-exact equality.

Imaging & DSP

Classic SC case studies encode pixels or filter coefficients as densities, then implement:

  • Local thresholding and morphological-style ops
  • FIR-like filtering with scaled MUX adds
  • Edge / texture features where small density errors vanish after quantization to display bits
graph LR
    Img[Pixels] --> SNG[SNG / B2S]
    SNG --> Kern[SC kernel: AND / MUX / counter]
    Kern --> Dec[Decode / binary out]
    Dec --> Metric[PSNR · SSIM · task score]

Design tip: compare energy at iso-PSNR (or iso-task), not iso-\(N\). A longer stream that wastes energy is not a win.

Coding & communications

Iterative soft decoders (LDPC, turbo-like BP, MIMO detection) already pass messages that are probabilities. Mapping those messages to stochastic wires can shrink check-node / variable-node logic, especially in fully parallel layouts.

Challenges unique to this domain:

Issue Why it matters
BER cliffs Small message bias can raise error floors
Throughput Frame latency vs stream length \(N\)
Mixed signal / CIM New MS-SCIM style datapaths need careful SNR budgeting

Shared lesson across both bridges

Applications succeed when designers treat SC as a noise channel with a known model, then allocate \(N\), parallelism, and hybrid binary stages the way a communications engineer allocates SNR.

Continue with taxonomies:

📚 DOI starting points