Streaming

Tick-by-tick transforms for bounded-window processing — five low-latency classical primitives at nanosecond cost, plus an adaptive family behind one shared incremental contract.

Low-latency primitives

Each primitive lives in ferro_wave::streaming and emits new coefficients per update(sample). State stays in a bounded ring buffer; steady-state allocation is zero.

Primitive	Description
`StreamingMODWT`	~67 ns per tick (Db4, 4 levels) — shift-invariant detail + scaling coefficients.
`StreamingDWT`	Decimated streaming DWT.
`StreamingSWT`	Stationary streaming transform.
`StreamingCWT`	100–600 ns per tick across 10–100 scales — a fresh (power, phase) column per sample.
`StreamingDenoiser`	42 ns p50 (Db4, 4 levels) — one denoised sample per tick.

use ferro_wave::streaming::StreamingMODWT;
use ferro_wave::wavelets::{Daubechies, DaubechiesType};
 
let wavelet = Box::new(Daubechies::new(DaubechiesType::Db4));
let mut stream = StreamingMODWT::<f64>::new(wavelet, window, 4);
 
for &tick in &feed {
    if let Some(update) = stream.update(tick)? {
        on_update(update); // per-tick coefficients
    }
}
# Ok::<(), ferro_wave::WaveletError>(())

EMD, EEMD, CEEMDAN, MEMD, VMD, SSWT, and EWT support fixed-window incremental use through the shared IncrementalTransform contract. The public contract covers readiness, bounded state, output shape, and error behavior. Refresh and reuse policies, solver initialization, cadence, thresholds, and related numerical controls are controlled implementation details.

Use it

Streaming denoise — emit one clean sample per tick with StreamingDenoiser.
Streaming CWT — update a Morlet scaleogram column on every sample.

See the sanitized streaming API reference for callable signatures and types.

Low-latency primitives

Adaptive transforms

Use it