Measure governed ingest-to-emit pipeline latency

The default benchmark processes 60 seconds of deterministic 32-channel, 512-Hz data in 32-sample chunks through the compiled causal bandpass plus rolling-RMS operation. Input generation and warmup are outside the measured interval. Results describe empirical performance on the recorded hardware, not a hard real-time guarantee.

Usage

measureLatency(
  pipeline = NULL,
  n_channels = 32L,
  sampling_rate = 512,
  hop_samples = 32L,
  duration_s = 60,
  warmup_chunks = 64L,
  seed = 1L
)

Arguments

pipeline

Optional empty StreamPipeline. NULL constructs the reference bandpass-plus-RMS graph.

n_channels

Exact positive channel count.

sampling_rate

Finite positive sampling rate.

hop_samples

Exact positive samples per chunk.

duration_s

Finite positive duration whose sample count is divisible by hop_samples.

warmup_chunks

Exact non-negative warmup chunk count.

seed

Exact non-negative local random seed.

Value

A plain benchmark result with summary, latency rows, pipeline state, hardware, configuration, and reference hash.

Examples

# Tiny deterministic benchmark; sampling_rate * duration_s must divide
# evenly by hop_samples.
bench <- measureLatency(n_channels = 2L, sampling_rate = 100,
                        hop_samples = 10L, duration_s = 0.5,
                        warmup_chunks = 1L, seed = 1L)
bench$summary$count
#> [1] 5