SALPA

Introduction

This document explains how to use the SALPA algorithm from the command line, or from within Python or Octave/Matlab. The algorithm itself is explained in Wagenaar and Potter, Real-time multi-channel stimulus artifact suppression by local curve fitting. J. Neurosci. Methods 120, 113–120, available from the Wagenaar Lab website.

Installation

Prerequisites:

  • SCons

  • A C++ compiler

  • Python (version 3.8 or later, optional)

  • Octave or Matlab (optional)

Download the source from github:

enter the download folder:

cd salpa

and build using Scons:

scons

This creates a single binary “build/salpa”, which may be copied to any convenient location on your $PATH.

Command line usage

The SALPA program reads and writes raw files comprising 16-bit signed integers, with arbitrary numbers of channels.

The program understands the following arguments:

  • -F f

    The sampling rate is f, measured in Hertz. This argument must precede any of the arguments that are measured in milliseconds. (Default: 30,000 Hz.)

  • -c n

    The number of electrode channels in the file is n. (Default: equal to the total number of channels (see -C), or 64 if neither is explicitly given.)

  • -C m

    The total number of channels in the file is m. As an example, MultiChannel Systems’ binary files contain 60 electrode channels and 4 auxiliary channels, for a total of 64 channel. Similar arrangements are found in some OpenEphys recordings. (Default: equal to the number of probe channels (see -c).)

    Auxiliary channels are copied verbatim from input to output, without any processing.

  • -t d

    The threshold on the residual error for resuming normal operation after detecting an artifact is d digital units. (No default; see -x.)

  • -x x

    The threshold on the residual error for resuming normal operation after detecting an artifact is x times the RMS noise of each respective channel. (Default: 3.)

    Note that -t and -x are mutually exclusive

  • -l τ

    (That’s lowercase ℓ, not upper case 𝐼.) The half-width of the template for artifact-shape fitting is τ, measured in milliseconds. (Default: 3 ms.)

  • -a t

    The window over which residual error is calculated is t, measured in milliseconds. (Can usually be left to its default value, 0.2 ms.)

  • -b t

    A segment of t milliseconds is blanked before resuming normal operation. t must be less than the τ value given with -l. (Can usually be left to its default value, 0.4 ms.)

  • -Z

    If -Z is given, “blanking” operations requested by -b are terminated early if the signal crosses zero.

  • -r a,*b*

    When the digital value of any channel dips is less than or equal to a, or greater than or equal to b, treat the following segment as artifact. (Default: a = -32767, b = 32767. See also -A.)

  • -A t

    Look ahead t milliseconds for automatic artifact detection. (Default: 0.2 ms. See also r.)

  • -f t

    Enforce a minimum duration of t milliseconds for automatically detected artifacts.

  • -P filename

    Artifacts are assumed to exist across all channels at the times specified in the named file. Each line of the file must contain two numbers: the start of a presumed artifact (in samples) and the duration of that artifact (also in samples). The lines must appear in temporal order.

  • -T n

    Use n CPU threads for processing. (Default: 8.)

  • -S n

    Use a buffer size of n scans. (Default: 4096, internally rounded down to a power of two.)

  • -M n

    Skip first n scans from the beginning of the file. (Default: do not skip.)

    Note: The numbers in the file specified with -P are measured after the skip. That is, if the number k appears in the file, the artifact is presumed to exist n + k from the beginning of the file.

  • -N n

    Only process the first n scans of the file. (Not including any skipped scans specified with -M; Default: process entire file.)

  • -B

    If given, subtract baseline before processing. Values specified with -r are relative to subtracted baseline.

  • -i filename

    Read input from the named file. (Default: read from stdin.)

  • -o filename

    Send output to the named file. (Default: write to stdout.)

Usage example

To process a Neuropixels file recorded by OpenEphys:

salpa -F 30000 -l 3 -c 384 -r-20000,20000 -i continuous.dat -o clean.dat

Python usage

A Python module named “salpa” is provided to wrap around the binary.

This module defines two functions: salpa.fsalpa.params and salpa.fsalpa.run.

TOC