Q auto-suggest

Recommend a Q-factor from sample data. Two modes: recover Q when you have both a float reference and the fixed-point integers (mirrors the Debugger's per-row Auto-Q button); pick Q when you only have a float signal and a target Word width.

Mode

Data type

Float samples

📂 Drop float values file (or click)

Fixed integers

📂 Drop fixed integers file (or click)

Plot

Float Fixed (recovered) Diff (×1)

How "recover" works

For each pair where the float reference is non-zero, compute ratio_i = fixed_i / float_i. Since fixed = float · 2^Q for the matching Q-format, the ratios cluster around 2^Q. Take the mean of the positive finite ratios, log₂ it, and round — that's your Q. The standard deviation across ratios indicates how confident the suggestion is (low σ ⇒ a single consistent Q produced the data).

How "pick" works

Take max|x| across the float samples. The smallest integer E with 2^E > max|x| is the number of integer bits needed to represent the signal without saturating the positive peak; Q = (N − 1) − E. The result is the tightest Q for that Word width — i.e. the most fractional precision you can afford.

← All tools