check
גלי מוח: קוגניציה, ביולוגיה ומה שביניהם - מפגש חמישי (15.4.20) - A Methodological Discussion on the Applications of Filters for Cleaning and/or Analyzing Time-Series Data | לימוד בחברותא

גלי מוח: קוגניציה, ביולוגיה ומה שביניהם - מפגש חמישי (15.4.20) - A Methodological Discussion on the Applications of Filters for Cleaning and/or Analyzing Time-Series Data

The meeting was held on 15/04/2020 using Zoom as previously, and was led by one of our group organizers, Nora Vrieler. Since the situation surrounding the COVID-19 pandemic has us working from home as much as possible many of us are working mostly on data analyses; we therefore thought this a good time to have a methodological discussion on the applications of filters for cleaning and/or analyzing time-series data.

As our guide for this discussion we chose the article “Filters: When, Why and How (Not) to Use Them” by Alain de Cheveigné and Israel Nelken (Neuron, 2019), as this article provides a good overview of the ways in which filtering might help or hurt the acquired raw data from a perspective common to all methods studying electrophysiological potentials, from EEG/MEG to single-cell recordings.

The main point that the article aims to bring across is that filtering, regardless of how exactly it is done, alters the raw data. Knowledge of the principles behind filtering helps us understand what kinds of side-effects can arise from filtering, and how to account for these in interpreting the filtered data.

Filtering frequencies from time-series data is usually depicted as a relatively simple operation taking place in the frequency-domain; however, this depiction usually neglects to account for information on oscillation phases. The article’s authors therefore take the time-domain perspective throughout, defining ‘filtering’ as a convolution of several (N) points on the raw input data x(t) with an impulse response function. This way it is easy to see that each point on the filtered output y(t) depends on raw data recorded at N time points, which means that recorded events are ‘smeared out’ in time by filtering. However, this is not the only way in which events can be affected: all filtering operations have some spurious effects near the boundary between the stop/pass frequencies (for example ‘ringing’ - the appearance of oscillations at frequencies not present in the input). Such effects are easy to overlook when thinking of the filtering operation solely in the frequency-domain but can be seen quite clearly from the filter’s impulse response function in the time-domain, where an ‘impulse’ is a short pulse that essentially contains all frequencies.

Following the explanation of these basic principles we discussed various examples of impulse response functions and their corresponding filtering operation, drawing both from the examples presented in the article and from our own experiences. Finally, we discussed another main recommendation of the article, which is to avoid using filters as much as possible. Indeed, there are often alternative strategies, such as ‘robust detrending’; or in the case of power line noise for example, care can be taken to repeat experiments at such intervals that the noise is averaged out over repetitions. It was also interesting to note the discrepancy between sampling frequencies commonly used in single-cell recordings (usually > 20kHz) and EEG recordings (often no more than 1kHz), as we realized that it says a lot about not only about the frequency-content of the signals we record, but also the time-base of the kinds of events we expect to see.

To end this meeting we briefly discussed the next meeting, which will be held in two weeks. Following this brief methodological diversion, we will next be returning to cognitive aspects of brain rhythms; specifically, Moran Aharoni will talk to us about temporal aspects of attention. And of course, we will be paying special attention to see that the reported results are not based on spurious effects!

 

 

Participants:

Moran Aharoni

Slav Pesin

Vitaly Lerner

Noa Rahamim

Gal Vishne

Nir Ofir

Nora Vrieler