HSA (Average) / HSA vs. Time

HSA stands for High-resolution Spectral Analysis. A conventional FFT spectrum always shows a tone as a broad peak with side skirts, because the analysis window itself is smeared across the frequency axis. HSA removes that smearing: it identifies the dominant spectral lines in each block and iteratively subtracts the known spectral fingerprint of the analysis window from them. The result is a spectrum in which tones appear as narrow, well-separated lines — even when they are far closer together than the FFT line spacing would normally allow.

Theory

Why FFT Peaks Are Broad

An FFT evaluates a finite block of samples. Multiplying the signal by a window function is equivalent to convolving the true spectrum with the spectrum of that window. Every real tone therefore appears as a copy of the window's frequency response centred on the tone frequency. Two tones that are closer together than roughly the width of that response merge into one peak, and a strong tone masks weak neighbours under its skirts.

Iterative Deconvolution

HSA reverses the convolution. For each block it repeatedly:

finds the strongest remaining spectral line and estimates its exact frequency and complex amplitude with sub-bin accuracy;
subtracts the window response scaled to that line from the spectrum;
stores the line and repeats on the residual.

After the requested number of iterations the extracted lines are placed on a finer frequency grid and rescaled, and whatever remains of the residual (the broadband part of the signal) is added back in. Tones therefore become sharp lines while the noise floor is preserved.

Resolution Enhancement

The extracted lines are written onto a frequency grid that is finer than the FFT grid by the HSA Resolution Enhancement factor. With the default block size of 4096 samples and an enhancement factor of 4, the spectrum is computed on a 16384-point grid, and the result contains

Number of lines = (Window Size × Resolution Enhancement) / 2 − 1 = 8191

frequency values. The DC and Nyquist values are not reported, so the first value lies one frequency step above 0 Hz.

HSA (Average) and HSA vs. Time

Both analyses share the same core. They differ only in what happens to the sequence of block spectra:

	HSA (Average)	HSA vs. Time
Result type	One spectrum per channel (2D curve)	Colour map, time × frequency (3D)
Block handling	Energetic (squared) average over all blocks	Every block is kept as one column
X axis	Frequency	Time
Typical use	Identifying stationary tones with maximum frequency accuracy	Following how those tones move during a run-up or a transient

Parameters

Parameter	Options / Range	Default	Description
Window Size	power of two, in samples	4096	Block length of the underlying transform. Longer blocks give better base resolution but poorer time resolution.
Overlap	0 – 100 %	50 %	Overlap between consecutive blocks. More overlap gives a smoother time axis at the cost of computation time.
Overlap Mode	Fixed	Fixed	Keeps the overlap constant for all blocks.
Window Function	Hanning, …	Hanning	Window applied to each block. HSA subtracts the response of exactly this window, so the choice affects how cleanly lines are separated.
Spectral Weighting	None, A, B, C, D	None	Frequency weighting applied to the resulting spectrum.
HSA Resolution Enhancement	integer factor	4	How much finer the output frequency grid is compared with the plain FFT grid.
HSA Iterations	integer	20	Number of lines extracted per block. Higher values resolve more simultaneous tones but increase computation time.
Suppress Background Spectrum	on / off	off	When enabled, only the extracted lines are shown and the broadband residual is discarded — a purely tonal view of the signal.
Frequency Resolution / Resolution Specification	value / Spectrum Size	1 / Spectrum Size	Selects whether the block length is entered directly (spectrum size) or derived from a target frequency resolution.
Order Limit, Min Order, Max Order	off / 0.5 / 20	off	Reserved for order-related evaluation, where the extraction is restricted to a range of orders of the rotational speed. Leave disabled for normal spectral analysis.

Iterations and Resolution Enhancement are the two parameters that really shape the result. If two tones still merge, first raise the iteration count; if the lines look quantised on the frequency axis, raise the resolution enhancement.

Output and Units

Acoustic channels — the spectrum is shown in dB(SPL), referred to 2×10⁻⁵ Pa. The footer single value is the total level obtained by summing the line energies.
Vibration channels — the channel unit is kept (for example m/s² or g) and the spectrum is displayed linearly as RMS amplitudes. The footer single value is the energetic sum a = √(Σ amplitude²).
Acoustic or vibration handling is derived from the channel unit; it can also be forced in the node properties.
Because the HSA lines are already scaled as equivalent sinusoidal RMS values, no additional window bandwidth correction is applied when the total level is formed.

Typical Applications

Separating closely spaced gear-mesh and shaft harmonics that an FFT merges into one peak.
Resolving electric-drive switching tones and their sidebands, where the sideband spacing is only a few hertz.
Measuring the exact frequency of a tonal complaint so it can be traced back to a rotating component.
Precise tonality and tone-to-noise work, where the accuracy of the tonal line frequency determines the result.
Following the movement of individual tonal lines through a run-up with HSA vs. Time.

Practical Tips

Run a normal FFT first to see the overall picture, then use HSA on the frequency region that matters. HSA is considerably more expensive than an FFT.
The signal must be reasonably stationary within one block. For fast run-ups, shorten the window size or use HSA vs. Time instead of HSA (Average).
Raise the number of iterations when many tones are present at the same time; too few iterations leave the weaker tones inside the residual noise.
Enable Suppress Background Spectrum to check whether a peak is genuinely tonal: if it survives, it was a line; if it disappears, it was broadband energy.
Use a cursor to read the exact line frequency; that value, divided by the shaft speed, gives the order of the component.

HSA is a line-extraction method, not a general-purpose spectrum. Broadband and strongly modulated signals are not made more accurate by it, and results should not be compared directly with FFT levels bin by bin.