power-spectral-density

Single-interval power spectral density (1-D PSD)

A guided version of the IDL SPEDAS pwrspc crib sheet: take one scalar channel over a stationary interval and produce the classic log–log PSD so you can read off the inertial-range slope (e.g. Kolmogorov ~f^-5/3) or a spectral peak. This is the single-interval spectrum — collapsed over time, no dynamic axis. For the time-evolving picture use the dynamic-power-spectrum / wavelet tools instead.

When to use

• "What's the PSD slope of |B| (or B_R, density, V) over this interval?"
• "Where's the spectral peak / break frequency for this window?"
• "Give me the quick-look FFT spectrum, I don't need the time evolution."

Tool chain (all already exist)

load_data_source → browse_data_parameters → fetch_data_product → a small local pwrspc(time, values) call → write {freq, power} to one .npz → render_tplot (single line panel, log–log), wrapped in create_spedas_analysis_bundle.

There is deliberately no MCP "PSD" tool: pwrspc returns plain arrays, so the spectrum is computed locally and only the resulting two columns are written to disk.

Backend (VERIFIED contract)

pyspedas.tplot_tools.tplot_math.pwrspc.pwrspc(time, quantity, noline=False, nohanning=False, bin=3, notperhz=False) returns a (freq, power) tuple of ndarrays DIRECTLY — it does not create or store a tplot variable, and it does not write a file. You call it in-process and capture the two arrays.

• time must be numeric Unix-second timestamps (a 1-D float array), not ISO strings / datetime64.
• quantity is the 1-D scalar values array, same length as time.
• noline=False removes a linear trend; nohanning=False applies a Hanning window; bin sets log-frequency bin smoothing; notperhz=False returns power per Hz.
• freq is in Hz (derived from the sampling interval implied by time), so irregular/non-monotonic timestamps corrupt the frequency axis — resample to a uniform grid first if cadence is irregular.