If classic amplitude training is the foundational method, (bipolar montages are the methodological detail that the foundational method's trainings often skip over).

Most neurofeedback practitioners come out of their first training able to set up a referential montage — active electrode at a clinical site, reference at the linked-ear or mastoid, ground at the forehead — and run a session. Most of those same practitioners, asked to set up a bipolar montage and explain what the signal actually represents in that configuration, give an answer that does not quite line up with what the amplifier is measuring. They know the cable arrangement (two actives, no separate reference for the channel). They know some of the sites where bipolar is used (C3-C4, F3-F4, T3-T4, sometimes F7-F8). They might know that neurologists often prefer a bipolar montage. But the deeper question — what is the bipolar signal a signal of, and what clinical question can it actually answer? — is not always one they have been asked to answer carefully.

This is not a complaint about practitioners. (At least not directly…) It is a complaint about how the field teaches the methodological vocabulary. Bipolar gets named in passing during initial trainings, gets used in protocols copied out of papers or out of vendor manuals, and gets folded into clinical practice without ever being given the dedicated attention it deserves (I’d say at least a half-day). The result is that a non-trivial number of clinical sessions are being run in bipolar configurations chosen for reasons that do not quite hold up.

I will give the topic as much of the half-day I can in written form here. But I want to be honest about where the post is going before we start, because the line I am going to argue for is sharper than the polite bipolar has its place version of the discussion. The thesis: strict bipolar protocols — recording in differential mode without a parallel referential record of each site — have a structural indeterminacy problem the field has stopped having to live with with the advent of modern computing. Any serious EEG system lets you record two referential channels and compute the bipolar derivative in software. (My opinion is black and white: there is absolutely no reason to still be recording EEG in a bipolar montage; you lose too much; record in two channel monopolar, then remontage. You get the same results while maintaining the specificity of the referential EEG. Honestly… even then, the situations where I reach for bipolar are narrow, principled, and never overlook the technical problems with the setup.)

That is the thesis. The next several sections will explain why.

A brief history

Bipolar recording predates neurofeedback by several decades. It came out of clinical EEG, where it was (and still is) the workhorse montage for seizure localization, focal abnormality detection, and a great deal of the routine EEG read that hospital neurology departments do.

The original logic was practical. If you arrange a chain of bipolar pairs across the scalp — a longitudinal banana montage running front-to-back, a transverse chain running ear-to-ear — you can localize the source of a focal abnormality by reading the phase reversal across the chain. A spike that appears with positive polarity in one pair and negative polarity in the next pair is sitting at the electrode between them. That kind of localization is much harder to do from a referential montage, where every channel shares a reference and the spatial information is distributed across the whole array. Clinical EEG built its diagnostic discipline on bipolar chains for that reason.

Neurofeedback inherited the bipolar montage from that clinical-EEG tradition, but used it differently. Sterman's early sensorimotor rhythm work, in the late 1960s and early 1970s, included bipolar C3-C4 protocols as one of the configurations explored alongside referential C3 and C4 recordings. The choice was partly practical (a single channel of differential recording on a two-channel amplifier was easier hardware to manage than running two separate referential channels), partly conceptual (the C3-C4 difference indexed something different from the C3 or C4 activity), and partly empirical (some of the early SMR results survived the bipolar configuration as well as the referential one).

By the time Joel Lubar's group was working on theta/beta protocols for attention deficits, in the late 1970s and through the 1980s, the bipolar option had become part of the standard practitioner toolkit. It was not the default — most ADHD theta/beta work then and now is referential — but it was available, and certain variants of the Lubar protocol explicitly used bipolar montages over central or frontal-central sites. (Why? I honestly don’t know. I wouldn’t do it, and I’ve never heard a convincing argument for it.)

The bipolar tradition got its second life through the frontal alpha asymmetry (FAA) literature, beginning in the 1980s with Richard Davidson's work on hemispheric specialization for affect, and through the 1990s and 2000s with Peter Rosenfeld's clinical translation of FAA findings into a neurofeedback protocol for depression. Frontal alpha asymmetry — the difference in alpha amplitude between left and right frontal cortex — is, almost by definition, a bipolar measure. The clinical protocol that emerged from that line of work uses an F4-F3 or F3-F4 bipolar configuration, reinforces a particular direction of asymmetry, and has accumulated a small but methodologically interesting evidence base in depression and certain anxiety presentations. (There are a number of proposed alternatives to training in this manner that seek to rebalance the same imbalance, but using “classic” amplitude training.)

Around the same time, the temporal-lobe practitioners — particularly those working with addiction, trauma, and certain complex anxiety presentations — found that the standard referential montage produced badly contaminated signals over the temporal poles (T3, T4, T7, T8). The reference (linked-ear, mastoid) was sitting too close to the temporal electrodes, and the small interelectrode distance left the signal vulnerable to picking up reference contamination, EMG from the jaw and neck, and other artifacts. Bipolar T3-T4 and related configurations offered cleaner temporal-lobe signals, and the temporal-lobe protocol tradition built around them. (I still don’t use it even in these cases. I prefer a Surface Laplacian or an Average Reference.)

Through all of this, bipolar stayed a montage choice within amplitude training rather than a separate methodological tradition. It does not have a Sterman, a Lubar, or a Birbaumer in the same sense. What it has is a set of clinical contexts where the bipolar configuration earns its place — and a much larger set of clinical situations where it gets used by inertia, or skipped over by inertia, without the choice being thought through.

Alternate names

The vocabulary around bipolar is muddier than it should be, partly because the term overlaps with several adjacent concepts in EEG analysis and partly because different vendors and lineages have used the word slightly differently. A short orientation:

• Bipolar montage. The cleanest name. Names the configuration directly — two active electrodes, with the signal being their differential potential — and is the term used most consistently in clinical EEG, in qEEG software, and in the academic literature.

• Differential montage, differential recording. Both technically accurate descriptions. Differential is the engineering term for what the amplifier is doing (subtracting the signal at one input from the signal at the other), and the term gets used interchangeably with bipolar in some training materials.

• Sequential montage. The term used in clinical EEG when the bipolar pairs are arranged in chains — Fp1-F3, F3-C3, C3-P3, P3-O1 — for the purpose of localizing focal abnormalities by phase reversal. Less common in neurofeedback, where bipolar pairs are usually used singly or in small numbers rather than in chains, but worth knowing if you read the clinical-EEG literature.

• Asymmetry training. Used most often for the frontal alpha asymmetry protocol — F4-F3 or F3-F4 — but sometimes generalized to any inter-hemispheric bipolar configuration (C3-C4, T3-T4, P3-P4). The term names the clinical use of the bipolar signal rather than the configuration itself.

• Inter-hemispheric training. Names for protocols that use bipolar pairs across the midline — left and right homologous sites — to train coordination or reduce asymmetry between hemispheres. Sometimes used as a synonym for asymmetry training; sometimes used for a slightly different clinical target (coordination rather than asymmetry correction). The distinction matters in practice and gets blurred in marketing.

• Common-mode rejection. Not a name for the montage but for what the differential amplifier is doing. A signal that appears at both inputs of a bipolar pair — mains noise at 50 Hz or 60 Hz, broad cortical activity that is the same at both sites — gets attenuated by the amplifier's subtraction operation. Common-mode rejection is the property that makes bipolar montages cleaner for some signals and impossible for others. Worth understanding even if you do not use the term in writing.

• Laplacian montage. Not the same thing as bipolar, but easy to confuse with it. A Laplacian (or small Laplacian, current source density) montage references each electrode against a weighted average of its immediate neighbors. The resulting signal indexes very local cortical activity in a different way than bipolar does — it is a spatial-derivative measure rather than a two-point difference. Some software packages let you switch between bipolar, referential, and Laplacian montages on the same recording. They are not interchangeable.

When you see a paper or a protocol describing a bipolar configuration, the first practical exercise — exactly as with classic amplitude training — is to translate the lineage label into the technical vocabulary. Frontal alpha asymmetry training is F3-F4 bipolar with a directional reinforcement on the alpha-band amplitude difference. Trans-hemispheric SMR might be C3-C4 bipolar with SMR-band amplitude reinforcement, or it might be C3 referential and C4 referential trained simultaneously — and these are not the same thing. Always check the montage. Always check the reference. The protocol name does not always tell you.

How the method works

Mechanically, a bipolar configuration is very simple. The equipment used is exactly the same as for a referential montage. Two active electrodes go on the scalp, placed at the two clinically chosen sites. Both connect to the same amplifier channel — one to the positive input, one to the negative input. A ground electrode is placed somewhere standard, typically on the earlobe or toward AFz. The amplifier computes the instantaneous difference between the two active inputs in real time, and that difference is the signal that the rest of the chain — filters, spectral analysis, threshold logic, feedback delivery — operates on.

There is no separate reference electrode for the channel in the way there is in a referential montage. The two active electrodes function as each other's reference. (You may sometimes hear practitioners describe this as referencing one channel against the other, which is loose but I guess not wrong.)

The clinical setup looks similar to a referential session in most respects. Same impedance check. Same session length (typically twenty to thirty minutes of active training). Same session frequency (two to three sessions a week in most clinical practices). Same software environment for the most part. Same range of feedback delivery options — game elements, video, audio tones, or some combination. The threshold logic — set the reinforcement threshold so the client gets feedback roughly 60–80% of the time in early training, then tighten as performance improves — is the same logic that applies to referential amplitude training.

What changes is what the signal is, and therefore what is being reinforced.

A few of the more common bipolar configurations in clinical use, with the clinical question each is set up to answer:

F3-F4 (or F4-F3) for frontal alpha asymmetry. The signal is the difference in alpha-band amplitude between left and right frontal cortex. The clinical target is shifting that asymmetry in a particular direction — most commonly, increasing right-greater-than-left frontal alpha (which corresponds to relatively reduced right-frontal cortical activation, given alpha's inverse relationship with cortical activation) in depression protocols following the Davidson-Rosenfeld lineage. The protocol's mechanistic claim is that shifting the frontal asymmetry shifts the approach-withdrawal balance the literature associates with depressive and anxious presentations.

C3-C4 for inter-hemispheric SMR or beta work. The signal is the difference in SMR or beta amplitude between left and right sensorimotor strips. The clinical target is sometimes increasing the difference (a hemispheric-specialization claim, less common), sometimes decreasing it (an inter-hemispheric coordination claim, more common in TBI and post-stroke work). The mechanistic story varies with the indication.

T3-T4 (or T7-T8 in modern 10-20 nomenclature) for temporal-lobe work. The signal is the difference in activity over the left and right temporal lobes, but the practical reason for the bipolar configuration is often signal hygiene as much as the differential measurement itself — the bipolar montage gives a cleaner temporal-lobe signal than referential montages over the same sites, because the reference contamination problem is reduced. Used in some addiction protocols, some trauma protocols, and certain complex anxiety presentations.

F7-F8 for frontal-asymmetry variants, Fp1-Fp2 for frontal-pole work, and P3-P4 or O1-O2 for posterior bipolar protocols. Each has its own clinical literature and clinical use; the underlying logic is the same — train a difference between two sites, in a configuration that gives the cleanest signal for the clinical question.

Within a session, the variables a practitioner adjusts are the same ones available in referential amplitude training — threshold tightness, reward and inhibit bands, session pacing, feedback type, the cognitive task the client engages with during training, transfer-task framing. The one extra variable, which is the whole point of this entry, is the interpretive frame the practitioner is holding the session under. A bipolar SMR session is not a referential SMR session with the reference moved; it is a differential SMR session, and that distinction shapes what the practitioner is watching, what the threshold is reinforcing, and what learning looks like from the client's side.

Mechanistic specifics

What is being trained, mechanistically, in a bipolar protocol?

At the level of the signal: the difference in band-amplitude (or in some protocols, the difference in slow-wave activity, or in some less common protocols, a phase relationship) between two cortical sites. The operant target is the dynamic of this difference over time — its magnitude, its direction, sometimes its variability.

That is mechanically distinct from referential amplitude training in several ways that matter clinically.

Common-mode rejection. Activity that is the same at both electrodes of a bipolar pair gets attenuated by the differential amplifier's subtraction operation. This is the whole reason mains noise drops out of bipolar recordings: the 50 Hz or 60 Hz signal from the line voltage appears almost identically at both electrodes, and the amplifier cancels it. But common-mode rejection is not limited to artifacts. It also cancels real cortical signal that happens to be present equally at both sites. If the clinical target is broad bilateral activation — say, increased alpha across both hemispheres in a relaxation protocol — bipolar will not see it. The very symmetry that the protocol is trying to reinforce makes the signal invisible to the montage.

This is the most underappreciated practical implication of bipolar configurations. The amplifier is selectively blind to whatever is bilaterally symmetric. That blindness is a feature when you want to focus on a difference (FAA work, inter-hemispheric coordination); it is a bug when you want to track something that is happening on both sides at once (broad relaxation, generalized sensorimotor inhibition, bilateral alpha enhancement).

Irreducible indeterminacy of the underlying activity. From the bipolar amplitude alone, you cannot reconstruct what either site is doing. Many distinct configurations of the two underlying-site amplitudes produce the same differential signal — when the protocol reinforces an increase in the bipolar differential, the brain might be raising the dominant site, lowering the recessive site, doing both, or shifting overall amplitudes while keeping the difference constant. The practitioner has no read on which. This is the central methodological reason strict bipolar configurations no longer make sense to me as a default clinical choice; the worked numerical version of the argument, and what it implies clinically, sit in Brendan's perspective.

Local-circuit sensitivity. The signal that survives the common-mode rejection is the activity that differs between the two sites. If the two electrodes are close together, what survives tends to be local-circuit activity that is spatially specific to one site or the other — exactly the localization logic that clinical EEG uses to find seizure foci. If the two electrodes are far apart (across the midline, for example), the surviving signal is the cross-region difference rather than the local-circuit specificity. The interpretive frame shifts with the inter-electrode distance.

Reduced reference contamination. In referential montages, the reference electrode is doing more than its name suggests. It is not a silent baseline; it is recording activity of its own, and that activity is mixed into every channel referenced to it. A linked-ear reference contaminates frontal channels with whatever the ears are picking up — temporal-lobe activity, jaw EMG, the residual cardiac signal that earlobes are notoriously prone to. Bipolar montages remove this contamination, at the cost of removing whatever common signal is shared by the two active electrodes. For sites where the referential contamination is severe (temporal poles, frontal poles in some setups), the trade is worth making.

Different artifact profile. Bipolar configurations are robust against some artifacts and vulnerable to others. Eye blinks, which appear strongly at frontal sites and decrease with distance from the eyes, are largely cancelled by bipolar pairs that are equidistant from the eyes (C3-C4, P3-P4). Mains noise is cancelled by all bipolar pairs to a degree referential montages cannot match. But movement artifacts that affect one electrode more than the other — head turning, cable swing, a loose electrode on one side — become more visible in bipolar than in referential, because the differential amplifier amplifies anything that is not common-mode. Practitioners reading a bipolar trace need to know which artifacts they have made smaller and which they have made larger.

Phase information is preserved differently. A bipolar pair is sensitive to phase relationships between the two sites in a way that referential montages are not. If the two sites are oscillating in phase, the differential signal cancels (low amplitude); if they are oscillating in anti-phase, the differential signal adds (high amplitude). The amplitude in the bipolar channel is not a measure of "how much" of the band is present at either site — it is a measure of how out-of-phase the two sites are at that band. This is one of the more subtle implications of bipolar configurations, and it is the one most often missed in clinical practice.

Operantly, the conditioning logic is the same as in referential amplitude training. The brain learns, gradually, to spend more time in the state where the reinforced amplitude condition (which, in bipolar, is a difference condition) is met. The state being recruited is no longer "elevated alpha at the central strip" or "elevated SMR over C3" but something more relational: "an asymmetric configuration where left frontal alpha exceeds right frontal alpha" or "a desynchronized configuration where C3 and C4 are out-of-phase in the SMR band." The reinforcement shapes the brain's willingness to enter and sustain that relational state, not an amplitude at any single location.

That mechanistic shift — from training activity to training relationship — is the conceptual handle that distinguishes bipolar from referential protocols, and it is the handle that most under-deployed trainings do not give their practitioners.

Overview of the science base

The science base for bipolar protocols is bifurcated. There is a well-developed literature on specific bipolar configurations for specific indications — frontal alpha asymmetry for depression is the clearest example — and there is a much larger but methodologically diffuse body of clinical work where bipolar montages are used inside broader amplitude-training protocols without being singled out for evaluation. Most of the standard ADHD theta/beta literature includes some studies that used bipolar configurations and some that used referential, without those subgroups being analyzed separately.

The cleanest body of evidence sits in the frontal alpha asymmetry tradition. The Davidson group's work in the 1980s and 1990s established the FAA construct — relatively greater right-than-left frontal alpha is associated with depressive presentations and with the withdrawal end of the approach-withdrawal motivational axis. The clinical-translation work that followed (Rosenfeld in the 1990s, then Allen, Harmon-Jones, and others) established that the FAA pattern can be modified through neurofeedback, and a series of studies through the 2000s and 2010s — Choi et al. on depression, Mennella et al. on emotion regulation, Wang et al. on anxiety — reported clinical effects consistent with the underlying model. Effect sizes were modest, designs were mixed, and the active-control problem was the standard one. More recent work — Cantisani and colleagues' analyses, Schaffer et al.'s replications, methodological reviews from Stewart's group — has raised questions about FAA's test-retest reliability and the generalizability of the early findings. The protocol still has a place; the place is narrower and more carefully scoped than the 2000s literature suggested.

The inter-hemispheric coordination protocols (C3-C4 SMR and beta work for TBI and post-stroke) and the temporal-lobe bipolar protocols (T3-T4 for addiction, trauma, and certain complex anxiety presentations) sit on thinner evidence bases. Both have plausible mechanistic stories. Both have clinical experience accumulated in specific clinics. Both have small controlled-trial literatures — partly because the relevant trials are expensive and hard to recruit for, partly because the field tends to lump bipolar protocols under broader amplitude-training categories in meta-analyses.

The diffuse evidence base — bipolar configurations buried inside broader amplitude-training studies for ADHD, anxiety, performance, sleep — is essentially the classic-amplitude-training evidence base seen through a different lens. The effects are not cleanly separable from the broader amplitude-training effects. Meta-analyses that lump referential and bipolar together recover the combined signal; meta-analyses that try to separate them rarely have enough studies in each subgroup for confident conclusions.

Two caveats specific to bipolar are worth flagging. First, montage description in published studies is uneven — a non-trivial number of papers describe their protocols with enough ambiguity that a reader cannot tell whether the montage was referential or bipolar. "C3 SMR training" could be either. Second, direct head-to-head comparisons of bipolar and referential versions of the same protocol, in the same population, are rare. Most of what the field knows about when bipolar is the right choice is clinical experience and mechanistic reasoning, not controlled empirical comparison.

The summary I would offer a colleague new to bipolar protocols: FAA has a real but contested evidence base; inter-hemispheric and temporal-lobe bipolar work have plausible mechanistic stories with thin trial literatures; and the methodological details that would let a reader distinguish bipolar from referential are under-reported across the field, which makes confident interpretation harder than it should be.

Strengths and weaknesses

Set out fairly, the bipolar configuration has the following profile:

Strengths

• Better signal hygiene in regions where referential montages are badly contaminated. Temporal poles and frontal poles are the clearest examples — bipolar pairs across these regions deliver cleaner signals than linked-ear or mastoid-referenced channels at the same sites.

• Excellent common-mode rejection of broad-band noise. Mains, slow drift, and other widespread artifacts drop out of the differential signal in ways referential montages cannot match without aggressive notch filtering.

• The signal answers a different clinical question — what is the difference between these two sites? — that referential montages cannot directly answer. For asymmetry protocols, inter-hemispheric coordination work, and any clinical target whose physiological correlate is a difference rather than an activity, bipolar is the right configuration on principled grounds, not just on convenience.

• Reduced sensitivity to bad reference electrodes. If the linked-ear reference is impedance-imbalanced or the mastoid electrode is loose, every referential channel suffers; the bipolar channels referencing each other are protected from that failure mode.

• Cleaner across some artifacts that affect both sites equally — symmetric eye blinks in equidistant frontal pairs, broad neck-shoulder EMG patterns. The artifact-attenuation comes for free, without filter design.

Weaknesses

• Irreducible indeterminacy — the headline weakness. The bipolar signal does not let the practitioner reconstruct what either underlying site is doing. With the two-channel referential alternative routinely available (see below), accepting this blindness as the cost of bipolar feedback is no longer a defensible default. Brendan's perspective walks through why.

• The signal is structurally less intuitive for the client. Single-site amplitudes have felt correlates clients can learn to recognize — sensorimotor quietude with SMR, sustained engagement with theta/beta, low-arousal interoceptive openness with alpha-theta. A bipolar differential has no comparable felt correlate. The result is that bipolar protocols are tilted toward operant conditioning alone, with the second active ingredient — conscious and voluntary self-regulation — largely unavailable. Brendan's perspective develops this; it is, in my view, the more important of the two weaknesses.

• Information loss in the other direction is also structural. Whatever is symmetric across the two sites is invisible to the differential signal. If the clinical target is broad bilateral activation, bipolar will not see it.

• The interpretive frame is demanding. Practitioners need to understand that the band-amplitude in a bipolar channel reflects phase relationships between the sites as much as power at either site. Most initial trainings do not cover this. The result is a non-trivial number of clinical sessions where the practitioner's mental model of the signal does not match what the amplifier is measuring.

• Protocol design is demanding. A bipolar pair has to be chosen, not defaulted into. Random pair choices produce signals whose interpretation is not clean.

• Some artifacts get amplified, not attenuated. Asymmetric artifacts — head movement that affects one electrode more than the other, a loose electrode on one side, unilateral jaw EMG — become more visible in bipolar than in referential, because the differential amplifier amplifies anything that is not common-mode.

• The published literature is uneven. Methodological details are under-reported, and a non-trivial fraction of papers describe their protocols with enough ambiguity that a reader cannot tell whether the montage was bipolar or referential. That is not the practitioner's fault; it makes evidence-based practice harder.

The modern alternative: two-channel referential with a software-derived bipolar

Every professional-grade system I would recommend — Thought Technology's ProComp lines with BioGraph Infiniti, Mitsar with WinEEG, and a few others — supports two-channel referential recording with on-the-fly software computation of a bipolar derivative. Place two electrodes at the two sites of interest. Reference each to a linked-ear or mastoid. Configure the software to display and feed back on all three signals: the two referential channels and the derivative.

Train on whichever signal the clinical question calls for. More importantly, watch what each site is doing while training on the differential. The indeterminacy goes away because you no longer have to reconstruct site activity from the differential alone — you have it in front of you.

Strict bipolar — no parallel referential record, only the differential — is the configuration to phase out. The differential measurement itself remains useful when the clinical question calls for it. The point is to stop giving up the underlying-site information when keeping it costs nothing.

Brendan's perspective

Two thoughts anchor this section. The first is the indeterminacy argument the rest of the article has been pointing at. The second is a deeper point about what bipolar protocols can and cannot teach a client to do.

The indeterminacy problem, and why strict bipolar should be obsolete

The bipolar configuration's mechanical core — the signal is the difference between two sites' activity — has a corollary the field has worked around for decades without naming. Once you collapse two channels into one differential channel, you lose the information you need to know what each site is doing. The differential value is consistent with an infinite family of underlying-site pairs.

Worked through in concrete numbers. A bipolar F3–F4 alpha-amplitude differential of 5 µV could be:

• (F3 = 10, F4 = 5). Left frontal alpha modestly elevated; right at a moderate-to-low level.

• (F3 = 5, F4 = 10). The asymmetry in the opposite direction. (If you are training on bipolar magnitude rather than the signed differential, this produces the same training signal as the first.)

• (F3 = 100, F4 = 95). Both sites saturated with alpha, small relative difference. The brain is in a dramatically different state from the first scenario; the bipolar signal is identical.

• (F3 = 95, F4 = 100). Same as the third, asymmetry flipped. Again, identical training signal.

Four very different physiological states. One reinforcement signal. The protocol is asking the client's brain to learn changes in a quantity that maps to infinitely many underlying configurations, and the practitioner cannot tell which one the client just landed in.

For decades, this indeterminacy was the cost of admission. Early hardware and software simply did not let you record two referential channels and compute the bipolar derivative in the same session. Bipolar earned its place because the alternative was worse.

That is no longer true. The two-channel referential setup with software-derived differential is now the default option on most professional systems. The practitioner can watch what F3 is doing, watch what F4 is doing, and see the F3–F4 difference at the same time. The indeterminacy disappears. You can still feed back on the differential if that is the clinical target — but you can read what is underneath it.

That is why strict bipolar should now be the obsolete option, not the standard one. The clinical questions bipolar protocols were once the only way to answer are now questions you can answer better with two-channel referential. When a tool is dominated — meaning, another tool does the same job better with no compensating loss — it stops being a live choice. The list of genuine exceptions is short, principled, and covered in Section 8.

The conscious-and-voluntary-self-regulation problem

Part 1 of this series describes neurofeedback's learning chassis as resting on two active ingredients: operant conditioning of brain activity, and conscious-and-voluntary self-regulation. Most clinicians hold the operant-conditioning piece firmly in mind and the self-regulation piece more loosely. The self-regulation piece is where a lot of the clinical durability lives. It is what lets the client carry the trained state outside the clinic, recognize it, and return to it deliberately. Without it, you are left with a state the client can produce on the equipment and cannot reliably find without it. The transfer-task discipline NeuroLogic builds into trainings — and that Barry and Barry+ are designed to support — is the operationalization of this. We do not just train the brain. We teach the client what they are training.

Conscious self-regulation depends on the client developing an internal sense of the state they are learning to enter. Single-site amplitude protocols lend themselves to this naturally. SMR amplitude correlates with sensorimotor quietude — physically still, sensorily quiet, cognitively engaged but not effortful. Theta/beta correlates with sustained cognitive engagement. Alpha-theta correlates with low-arousal interoceptive openness. In each case, the signal at the site has a felt correlate, and that correlate is what the self-regulation lever grips onto.

A bipolar differential does not have an accessible felt correlate. The difference between left and right frontal alpha is not a state. The client cannot tell whether they are shifting toward more left frontal alpha or less right frontal alpha — both produce the same reinforcement, and neither easily maps onto a recognizable internal experience. The signal is real and physiologically meaningful. It just does not show up in interoception. (Think about it… I can have really high but balanced frontal alpha and be symmetric, just as I can have really low but balanced alpha and get the same result. The difference in frontal amplitude will be what drives the internal state change, not the left/right balance.)

The practical consequence is that bipolar protocols are structurally tilted toward operant conditioning alone. The self-regulation lever is largely unavailable. The brain can be conditioned to spend more time in a configuration that produces the reinforced differential, but the client is not learning the state in a way that survives the equipment. In most populations, that is a real clinical cost — and the second reason I think the field's enthusiasm for bipolar protocols has historically been overcalibrated relative to what they actually deliver.

There is one population where the inaccessibility becomes the clinical lever rather than the clinical cost. Section 8 covers it.

One additional note

Marketing language around bipolar is mostly fine, with one exception. The exception is trans-hemispheric used as a generic term to imply that any bipolar pair across the midline is doing the same kind of work. C3-C4 bipolar SMR training and F3-F4 bipolar FAA training are both trans-hemispheric in the literal sense. They are answering very different clinical questions and recruiting very different physiological states. The marketing flattens distinctions that the clinical work does not.

The thread across all of this, the one to leave the reader with, is the same one from the classic amplitude training entry: the method matters less than the assessment and the interpretive discipline layered on top of it. Bipolar versus referential is one of the cleanest tests of that discipline a practitioner faces. The choice is small in the moment. It is large in implication, because it determines what signal the rest of the protocol will reinforce — and what the client can or cannot learn from the session.

Would I do this method myself? In what context?

Yes — but the conditions are narrow, the hardware setup matters, and the clinical population is specific.

The hardware condition is non-negotiable. I no longer use strict bipolar configurations in clinical practice. When the clinical question calls for a differential measurement, I configure two referential channels at the two sites of interest and let the software compute the bipolar derivative. I feed back on the differential where the protocol calls for it. I keep the referential channels visible during the session so I can read what each underlying site is actually doing. Any answer to would I do this? that begins with yes, in strict bipolar configuration — no parallel referential record — is one I am no longer willing to give.

The clinical-population condition is where the yes in this section earns its place.

There is a population for whom the non-intuitiveness of the bipolar signal stops being a clinical cost and starts being a clinical lever. These are the clients for whom conscious control is the problem rather than the solution. Perfectionists. Hypervigilant clients. People whose attempts to fix themselves through effort have produced exactly the rigidity they came in to undo. Clients who tighten when they try to relax, scatter when they try to focus, brace when they try to let go. The familiar paradox: the thing the client is trying to control is being maintained by the trying.

For these clients, a single-site amplitude protocol can quietly backfire. The felt-state correlate of the signal — the feature that makes amplitude protocols good at engaging conscious self-regulation in most populations — becomes another handle for the conscious-control system to grip onto. They start trying to make the bar go up. They notice when they fail. They redouble the effort, and the effort itself moves them away from the state the protocol is meant to reinforce. The interoceptive accessibility of the signal becomes a vehicle for the very pattern the client is in the room to unlearn.

Bipolar-derived feedback does something useful in exactly these cases. The signal is opaque to the conscious-control system. There is no story about what I should do internally to make the bar go up. The only way to influence the signal is to step back from the trying. The brain finds the reinforced state slowly, through operant conditioning, without conscious effort being able to interfere in the same way. The non-intuitiveness becomes the feature. Conditioning without conscious interference is exactly what these clients need a path to.

This is a real clinical use case, and it is the one place I will reliably reach for bipolar-derived feedback. It is also narrower than the field's every method has its place framing suggests. Outside this population — and outside a handful of signal-hygiene edge cases in regions where referential montages are genuinely too compromised to use — referential training with appropriate transfer-task discipline is the better answer for the great majority of clinical questions bipolar protocols used to be reached for.

What I would say to a colleague starting out, and what I say in NeuroLogic's trainings: spend a deliberate half-day learning what the bipolar signal actually represents. Learn to configure two-channel referential setups with software-derived differentials so the indeterminacy stops being something you have to live with. Reserve bipolar-derived feedback for the narrow population where its non-intuitiveness is part of the clinical work. And articulate, every time, why this configuration, for this client, with this formulation. The gap between what you think you are training and what you are actually training is widest here. The discipline of closing it is what separates deliberate practice from menu-driven practice — in this method and in every other one in the series.

The next post — Part 4 — Coherence and connectivity training — turns to the methodological family that explicitly trains relationships between sites rather than activity at sites. Bipolar is the simplest version of that family — a two-point, single-band coordination measure. Connectivity training scales the same conceptual move up to network-level metrics, with similar interpretive demands and several of the same indeterminacy issues. The reasoning that earns bipolar a narrow, principled place in clinical practice is the reasoning the next entry will need to apply at greater scale.

References

• Allen, J. J. B., Coan, J. A., & Nazarian, M. (2004). Issues and assumptions on the road from raw signals to metrics of frontal EEG asymmetry in emotion. Biological Psychology, 67(1–2), 183–218.

• Cantisani, A., Koenig, T., Stegmayer, K., Federspiel, A., Horn, H., Müller, T. J., … Walther, S. (2016). EEG marker of inhibitory brain activity correlates with resting-state cerebral blood flow in the reward system in major depression. European Archives of Psychiatry and Clinical Neuroscience, 266(8), 755–764.

• Choi, S. W., Chi, S. E., Chung, S. Y., Kim, J. W., Ahn, C. Y., & Kim, H. T. (2011). Is alpha wave neurofeedback effective with randomized clinical trials in depression? A pilot study. Neuropsychobiology, 63(1), 43–51.

• Davidson, R. J. (1992). Anterior cerebral asymmetry and the nature of emotion. Brain and Cognition, 20(1), 125–151.

• Harmon-Jones, E., & Allen, J. J. B. (1998). Anger and frontal brain activity: EEG asymmetry consistent with approach motivation despite negative affective valence. Journal of Personality and Social Psychology, 74(5), 1310–1316.

• Mennella, R., Patron, E., & Palomba, D. (2017). Frontal alpha asymmetry neurofeedback for the reduction of negative affect and anxiety. Behaviour Research and Therapy, 92, 32–40.

• Rosenfeld, J. P., Cha, G., Blair, T., & Gotlib, I. H. (1995). Operant (biofeedback) control of left-right frontal alpha power differences: Potential neurotherapy for affective disorders. Biofeedback and Self-Regulation, 20(3), 241–258.

• Schaffer, C. E., Davidson, R. J., & Saron, C. (1983). Frontal and parietal electroencephalogram asymmetry in depressed and nondepressed subjects. Biological Psychiatry, 18(7), 753–762.

• Sterman, M. B., & Friar, L. (1972). Suppression of seizures in an epileptic following sensorimotor EEG feedback training. Electroencephalography and Clinical Neurophysiology, 33(1), 89–95.

• Wang, J. R., & Hsieh, S. (2013). Neurofeedback training improves attention and working memory performance. Clinical Neurophysiology, 124(12), 2406–2420.