Reply to thread

Message: [QUOTE="Bombastus, post: 92064, member: 2178"] Prismate’s spectrum analyzer uses a 2048-point FFT, refreshes at 30 Hz, and plots frequency on a logarithmic axis. Those numbers sound impressively specific, but they describe three different parts of the display rather than one measure of accuracy. FFT size controls how much audio enters each transform and how closely its frequency bins are spaced. Refresh rate tells you how often the graph redraws. A logarithmic axis changes how those bins are arranged visually across the screen, not how many measurements the transform actually produces. [HEADING=2]FFT size trades low-end detail for faster movement[/HEADING] At 48 kHz, a 2048-point FFT spaces raw bins 23.4375 Hz apart because the sample rate is divided by the transform size. Each transform also spans about 42.7 milliseconds of audio. At 44.1 kHz, the same FFT gives roughly 21.5 Hz spacing across a 46.4 millisecond block. This is the basic time-frequency trade-off behind any real-time audio spectrum analyzer. Doubling the FFT to 4096 halves the bin spacing, but it also doubles the block length. The graph can separate finer spectral structure, yet sudden changes become less tightly located in time. The low end exposes the compromise quickly. At 48 kHz, a 2048-point transform supplies only four positive-frequency bin centers below 100 Hz, at roughly 23.4, 46.9, 70.3, and 93.8 Hz. A smooth curve drawn between them can look richly detailed, but the underlying transform has not secretly created extra bass measurements. This is why [B][URL='https://goldmidi.com/community/threads/remi-blaze-has-introduced-free-prismate-eq-for-macos.77440/']Prismate’s 2048-point spectrum analyzer[/URL][/B] is better for broad visual guidance than microscope work in the sub-bass. You can still see where low-frequency energy is collecting, follow changes while EQing, and spot obvious tonal imbalance. Trying to read a tiny difference between two nearby bass components directly from the graph asks more from the raw FFT than its size provides. [HEADING=2]Windowing changes what frequency resolution really means[/HEADING] Bin spacing is useful, but calling it the analyzer’s exact resolution is too tidy. Real audio rarely fits a transform block as an integer number of cycles, so energy from one component can spread into neighboring bins. Window functions reduce that leakage by tapering the captured block, although the taper also broadens the main lobe around a spectral peak. The classic [B][URL='https://ieeexplore.ieee.org/document/1455106/']windowing trade-offs in harmonic analysis[/URL][/B] show why FFT size alone cannot tell you whether two nearby components will appear as separate peaks. Different windows trade main-lobe width, sidelobe suppression, and amplitude accuracy differently. Prismate’s public specification gives the FFT size but does not identify its window, so its exact close-frequency resolving behavior cannot be calculated from “2048-point” alone. There is another wrinkle. Peak interpolation can estimate the frequency of one clean, stable tone between FFT bin centers, so a 23.4 Hz bin spacing does not mean every frequency estimate must jump in 23.4 Hz steps. Separating two simultaneous tones is a different problem from estimating the center of one isolated peak. Zero padding does not rescue the underlying separation either. Adding zeros before the transform can provide more plotted sample points and make a spectrum look smoother, but it does not add new captured audio. More genuine frequency discrimination requires a longer observation window, a different estimation method, or both. [HEADING=2]A logarithmic graph can make bass look more precise[/HEADING] Human hearing and musical pitch are usually easier to inspect on a logarithmic frequency axis because octaves receive comparable visual space. Twenty to 40 Hz occupies the same horizontal distance as 10 to 20 kHz on a true logarithmic frequency axis. The interface becomes much easier to read than a linear plot where most musical bass is crushed against the left edge. The display choice has a side effect. A logarithmic axis stretches the sparse low-frequency FFT bins across more pixels while squeezing thousands of higher-frequency bins into less space. A beautifully fluid bass trace can therefore look more precise than the underlying low-frequency samples really are, especially if the interface also interpolates or smooths the line. Prismate’s 30 Hz refresh rate belongs to a separate part of the story. Thirty redraws per second means the visual can react quickly enough to feel live, but it does not shrink the 2048-point bin spacing. The developer does not publish the analyzer’s hop size or overlap, so the refresh figure should not be used to infer how each FFT frame is assembled. For everyday EQ work, none of this makes a 2048-point analyzer inadequate. Fast visual feedback is useful when you are watching a vocal brighten, checking whether a kick carries excessive low-mid energy, or confirming that a shelf is moving the expected region. For exact sub-bass measurement, closely spaced tones, or forensic resonance work, a larger FFT or dedicated analyzer gives the low end more actual data instead of merely more screen space. [/QUOTE]

Name