IndieRF Phase Noise to Jitter Calculator
Phase noise to jitter calculator
A 100 MHz clock falling 10 dB per decade from −120 dBc/Hz at 1 kHz to a −150 dBc/Hz floor has 344 fs rms jitter from 12 kHz to 20 MHz, which caps a 70 MHz ADC input at 76.4 dB SNR. Type or paste your mask, set the carrier and the band, and read rms jitter, rms phase error and the SNR limit, with the band shaded on the plot.
Phase noise to jitter
100 MHz carrier, integrated from 12 kHz to 20 MHz
Phase noise mask
Single-sideband L(f) in dBc/Hz at each offset, as the datasheet prints it. Offsets take Hz, k, M or G (10k, 1.5M). Points are joined by straight lines on the log-log plot.
Where the jitter comes from
In the example the floor from 1 MHz to the top of the band carries 81% of the phase variance, because a −150 dBc/Hz floor across 19 MHz outweighs the close-in noise that looks worse on the plot. Starting the band at 1 kHz instead of 12 kHz only moves the jitter from 344 fs to 362 fs. Here a cleaner floor or a narrower band buys more than a cleaner close-in. Read the upper limit from what follows the clock: the ADC's input bandwidth, or the loop bandwidth of the PLL it feeds.
The SNR limit is set by the jitter and the analog input frequency, and the sample rate does not enter it. Each doubling of the input frequency or of the jitter costs 6 dB. It counts this clock alone: the ADC's own aperture jitter adds in root-sum-square, so put the total in when the datasheet gives one.
Definitions
- L(f) is single-sideband phase noise in dBc/Hz at offset f. Between two points the page draws a straight line on the log-log plot, so each segment is a power law and integrates in closed form.
- RMS phase error counts both sidebands: φ²rms = 2 ∫ 10^(L(f)/10) df over the band, in rad².
- RMS jitter: σt = φrms / (2π fc), with fc the carrier.
- ADC SNR limit: SNR = −20 log10(2π fin σt), with fin the analog input frequency.
- EVM floor: φrms in radians, as a percent, the small-angle value.
- A band that runs past the first or last mask point is refused, with the reason shown. Extending the mask would mean guessing the slope, so add a point instead.
The math runs in your browser and the mask goes into the link, so a shared link opens with the same points. The Cascade analysis page runs the same integral on the LO of a mixer in your chain and adds reciprocal mixing from a blocker.
Checked against scipy numerical integration: the cases, the tolerance and what the check does not cover.