Streaming denoise

StreamingDenoiser emits one denoised sample per update(tick) from a bounded ring of MODWT coefficients — measured at 42 ns p50 on a single core for Db4 at 4 levels, with zero steady-state allocation.

When to use it

• You are on a live feed and cannot afford to re-batch a window on every update.
• You want a denoised value emitted per tick for a strategy or online-inference loop, at nanosecond latency.
• You need bounded, predictable memory — state lives in a ring buffer, not a growing history.

Example

use ferro_wave::{Daubechies, DaubechiesType, BoundaryMode};
use ferro_wave::streaming::denoising::{StreamingDenoiser,
    StreamingDenoiseConfig, ThresholdMethod, ThresholdRule};
 
let wavelet = Daubechies::new(DaubechiesType::Db4);
let config: StreamingDenoiseConfig<f64> = StreamingDenoiseConfig {
    levels: 4,
    method: ThresholdMethod::Soft,
    rule:   ThresholdRule::Universal,
    boundary: BoundaryMode::Periodic,
    ..StreamingDenoiseConfig::default()
};
 
let mut denoiser = StreamingDenoiser::new(&wavelet, config)?;
 
for &tick in &live_feed {
    let out = denoiser.update(tick)?;
    on_clean(out.clean); // emit to strategy
}
# Ok::<(), ferro_wave::WaveletError>(())

Notes

• Per-tick cost is O(filter_length × levels); steady-state allocation is zero after warm-up.
• The configuration mirrors batch denoise — same methods, rules, and boundary modes — so the streaming path tracks the batch result.

Use the sanitized streaming API for the current denoising contract and update-trigger configuration.