Noise blankers

Linrad removes impulse noise with two blankers, both operating on the weak-signal part of the timf2 time function, where the strong signals have already been split off (see First FFT and the strong/weak channel split):

• a clever blanker that detects each impulse, measures its shape and — in a two-channel system — its polarization, fits a stored reference pulse, and subtracts it, leaving the weak signal underneath intact; and
• a stupid blanker that zeroes every sample above a threshold plus a guard region. It is destructive to any weak signal under the pulse and is used as a fallback, and as the only blanker when the clever one is disabled.

The two blankers use the code's own names, clever_bln and stupid_bln.

What runs each cycle#

Once per timf2 update the blanker processes the samples between where it left off (timf2p_fit) and a point a safety margin behind the newest data (so a pulse straddling the boundary is fitted next time). It runs the clever blanker first, then the stupid blanker over the same region, then updates the noise-floor statistics that set both thresholds.

Clever blanker: detect, fit, subtract#

The clever blanker removes each impulse by matched subtraction rather than gating. The flow for each detected pulse:

It scans the per-sample power for a peak above clever_bln_limit, checks the peak is an isolated single pulse of a plausible width, and, for a genuine pulse, determines how to reproduce it: the phase is de-rotated to a reference, the sub-sample peak position is found by a parabolic fit, and that fractional position selects one of 256 stored reference-pulse shapes (the shape of a band-limited impulse depends on where its peak falls between samples). The scaled reference is subtracted from timf2 and the per-sample power rebuilt.

If the residual power in the subtracted region exceeds half the original, the model did not fit: the original data is restored and the pulse is flagged so it is not retried. A good fit is flagged as handled, and the scan continues.

Pulse polarization (two channels)#

With two RF channels the pulse is first collapsed onto a single polarization. Assuming one common source with equal, uncorrelated noise in both channels, the normalized 2×2 covariance of the two channels over the pulse gives the polarization angle and the combining coefficients C1, C2, C3; the reference pulse is subtracted from both channels using those coefficients. This is the same computation used to receive the wanted signal on its optimum polarization (see Diversity and adaptive polarization), applied here to cancel an interference source.

Stupid blanker: threshold clear#

The stupid blanker zeroes any run of samples whose power exceeds stupid_bln_limit. Around each cleared run it also clears a guard band whose width grows with pulse strength, capped at the equivalent of about 40 dB.

Noise-floor tracking and adaptive thresholds#

Both thresholds derive from a continuously updated noise-floor estimate, measured only from the samples the blanker did not touch. In automatic mode the limits are the noise floor times a user factor. If the stupid blanker is clearing more than ~20 % of samples (a busy band rather than impulse noise), the noise-floor estimate is raised to back the blanker off. The blanker rates are reported on screen.

Parameters#

From the HG_PARMS hg struct (globdef.h:972):

Related: strong/weak channel split · diversity and adaptive polarization · signal-chain overview