The High-Pass Filter Hypothesis for Autism Spectrum Disorders

A high-pass perceptual filter influences sensorial perception and thus causes granular categorisation and a resistance to integrate conflicting interpretations of the world.

I hypothesise that the cognitive properties of the Autism Spectrum Disorders can be caused by the presence of a high-pass filter applied to the sensorial perception. By high-pass filter, I mean a filter which enhances the fine details and lessens the broader patterns (I do not use the term with its electronic meaning of “filter which blocks low frequencies). The more one person is deep on the Autism Spectrum, the stronger the effect of such filter. For a figured representation of how the filter affects one’s perception of the world, please see my article “The World Through a Magnifying Glass”.

The fact that this filters relays to perception a preponderance of details causes the people with ASD to organise concepts in narrower categories with sharper boundaries.

Due to the more granular and more strictly defined categories, concepts appear more distant and thus exert a lower reciprocal inhibition. Hence, a person with ASD tends to maintain in his mind a higher number of potentially contrasting concepts, instead of collapsing or integrating them in a single interpretation. The presence of contrasting thoughts and their slower integration is responsible for a series of behaviors observed in ASDs.

I suspect that the impairments in integration cause a person with ASD to have a slower or weaker emotional response to affective behaviour demonstrated by others, and thus in producing lessened affective behaviour in the ASD person as well. This impairment in receiving or sending emotional messages might be responsible for many of the social deficits expressed by ASDs.

I will now discuss in more detail the concepts I just described.

The high-pass filter

The general formulation of a high-pass filter is “a filter which lets narrow and local changes pass through and blocks broader patterns”. Such a perceptual filter (whose exact mechanism is yet unknown) causes people to better perceive local variations (“details”) at the expense of less local features (“context”).
Some experimental results point in the direction that rather to have an impaired global processing, autistic people would have a stronger local processing, which would tend to take over the global one. In such case, to be formally correct, the definition of high-pass filter for the purpose of this hypothesis, would be: “a filter which applies a gain to local changes and lets unchanged broader ones”.

I postulate that each person, including neurotypical ones, possesses a high-pass filter. However, its intensity (how much a small detail is enhanced) and its bandpass (how small a detail has to be in order to be enhanced) are the qualities that determine whether, and how far, a person sits on the Autism Spectrum.

The high-pass filter causes a different perception, as I try to express in my article “The World Through a Magnifying Glass”. The high-pass filter does not act (only) on features which possess a frequency, but works on many dimensions: visual detail, colors, object features, word meanings, and so on. The best way to imagine the effect of an high-pass filter is a filter which examines each little piece of perception and compares it with the adjacent ones: if it is different on any salient dimension, it applies a gain factor to the perceptual piece. However, this comes at the expense of missing out on global patterns, i.e. changes which are distributed on more adjacent perceptual pieces. Please see the example below. The first line is a random sequence of numbers. The second line represents the effect of filtering the first through a high-pass filter, whose equation is:

The high-pass filter equation
The input and output sequences
A graph of the two output sequences

Observing the orange line only in the graph above (the sequence passed through the low-pass filter) we perceive a general trend of an increase in the first part of the sequence and a gradual decrease in the second part; observing the blue line (the sequence passed through the high-pass filter) we instead perceive a single bump in the middle of the sequence (greater perception of detail) but no gradual increase (lesser perception of trend).
The high-pass filter caused us to focus on a local point (the bump) while underestimating the broader features.


A high-pass perceptual filter leads to narrower categories with sharper boundaries. Intuitively, the more details are attended to in an object, the bigger will be the absolute distance from existing categories and the higher the chance that a new category is formed. A more formal explanation is that given the “size” of an object, the space formed by the possible low-frequency configurations must be smaller than the space formed by the possible high-frequency configurations; for a random set of objects the probability to be close in the low-f-space space is higher than in the high-f-space (because is smaller); thus, clustering with a fixed distance threshold, the expected number of clusters is higher in the second space.

The narrow categorization caused by the high-pass filter likely causes also an increased noise sensitivity, as predicted by the Hierarchical Temporary Model of the human cortex. Given a cortex region responsible for discriminating a sensorial input, increasing the number of categories while mantaining the number of neurons fixed causes a decrease in redundancy, which itself is what prevents noise from influencing the results of the computational work performed by the cortex region.


The fact that categories describing different interpretations of a perceptual feature or situation are more distant from each other, due to the narrower categorization, reduces the reciprocal inhibition. There are many mechanisms in the brain which cause reciprocal inhibition, like the ones occurring in minicolumns and in the basal ganglia, in which the option faster to reach a certain threshold inhibits the slower ones. The high pass hypothesis, which implies a narrower categorization, predicts that people on the Autistic Spectrum have an increased tendency to have different interpretations of a perceptual item or situation coexist at the same time in their minds.

Let me use a thought experiment. Think about an acquaintance of yours. Is she a good person? You would probably remember a few situation where she behaved like a good one and a few where she didn’t; your mind will weight them and produce a single answer like “she is a good person” or “she is ok”. What happened? The “good experiences” started inhibiting the bad ones and vice versa; the side with the most and strongest experiences inhibits the other (while being partially inhibited itself) and all the experiences you had with your acquaintance collapsed into a single judgement. A possible consequence of the High-Pass Filter hypothesis is that such reciprocal inhibition mechanism is slightly impaired in the Autism Spectrum which leads to a higher tendency to maintain contrasting interpretations of reality in the autistic mind.

Actually, our brain is wired to maintain such contrasting representations: it is the trademark of parallel computing. The brain holds true different hypothesis or options, forwards them all to the next computing modules, performs parallel computations, and only at the end decides which option to pursue or pass to consciousness as a perception of reality. Intuitively, that’s what we sense happening in our mind when we are undecided on alternatives and go back and forth between them. Maintaining contrasting representation is a phenomenon that happens in everyone; I postulate only a bit more in the Autism Spectrum.

The idea that contrasting representation coexist until collapsed to produce a judgement to pass to consciousness has already been explored in Quantum Mind theories and well explains some experimental results through destructive interference (though I am skeptical on some of the QMT points: the fact that two systems share a phenomenon doesn’t mean they follow the same rules). It would be interesting to see if a future study can confirm that such phenomenon happen with a different strenght in the autistic mind. I propose that this, in addition to the high-pass filter effect which produces cogntitive styles like the Weak Central Coherence, explains some of the experimental results where people with ASD perform well like the embedded figures test: they can maintain in their conscious mind different interpretations of what they perceive.


Emotions happen in concurrence to strong judgements. When the autistic mind produces multiple and discording judgements, emotions will be weaker or less defined.

Let’s continue the previous thought experiment. I see an acquaintance walking towards me and I remember the experiences we shared, good and bad ones. A neurotypical mind would tend to collapse those experiences in a single judgement, from which a clear emotion would arise, which would in turn prime the next behaviors. For example, it would decide that the acquaintance is a good person, joy would be felt and expressed with a big smile. Conversely, in the same case an autistic mind would produce two judgements: “I like her” and “I don’t like her”, maybe with the former stronger than the latter. This would not be enough to feel a clear emotion and no social gesture as of a big smile would be adopted (of course I am exaggerating the phenomenon here just to provide the reader with an easier comprehension of it; in reality the difference between the two reactions is much weaker).

Social interaction is based on collapsing impressions to produce fast and clear responses. The reticence of individuals with ASD to fully make such integration could be one of the contributing factors to their social impairements, as slow and mixed social messages are not well received by neurotypicals, who prefer a social communicative style close to theirs.

Lessened reciprocal inhibition could also be one of the contributing factors to the slower task switching often recorded in people with ASD, as they have impairments in fully suppressing the old task in favor of the new one.

This article is a work-in-progress piece: a review of how this integrates with previous theories, of its compatibility with experimental results and how it predicts most symptoms of ASD will be added in the following days or weeks. Meanwhile you are welcome to leave your comments, here on Medium or by sending me an e-mail at You can also follow me on Twitter: @DellAnnaLuca.

To receive a notification when I publish a new post on the topic, please click here.