{"id":"86862020-cecf-4048-917c-3d74f59cff02","arxiv_id":"2608.06783","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A single-microcontroller wearable platform records 8-channel EEG at up to 8 kHz while generating tDCS, tACS, and temporal interference stimulation, with validation on bench tests and a gelatine head phantom.","lead":"This paper reports a compact wearable device that records 8-channel brain signals at up to 8,000 samples per second while also delivering electrical stimulation, including temporal interference stimulation, from a single microcontroller. Bench and head-phantom tests showed high signal fidelity and stimulation current errors below 1%, with the device capturing mixed brain-signal and stimulation waveforms without clipping.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Concurrent 'without saturation' claim is not reproducible because the ADS1299 gain setting for Figure 9 is unspecified; the stated ±375 mV dynamic range cannot accommodate the volt-level artifacts the paper itself reports.","rationale":"The reader's weakest-assumption analysis correctly identifies the unspecified amplifier configuration as the key gap. The paper's own text creates the tension: a ±375 mV dynamic range is stated, and volt-level artifacts are then described. Without knowing the gain used in Figure 9, an independent reviewer cannot verify the 'without saturation' claim, nor can they assess whether the simultaneous EEG channel still has enough resolution to preserve EEG signals. This is not a fatal flaw, because the missing information is straightforward to supply and the benchtop stimulation accuracy results are self-contained and plausible. The verdict should remain CONDITIONAL: the paper needs a clarifying revision reporting the gain/dynamic-range settings and ideally a quantitative EEG-fidelity test under concurrent stimulation. The concern is precisely the one the reader raised, so agreement is 'agree'.","tokens_in":15870,"tokens_out":3369,"duration_ms":39607,"concrete_test":"Ask the authors to report the ADS1299 PGA gain, reference voltage, and channel input configuration used in the EEG-only phantom trial (§III-C-1) and the concurrent trials (§III-C-2), including the effective input-referred noise and LSB size at those settings. Then repeat the concurrent phantom experiment with a calibrated 10 µV, 10 Hz EEG template injected during 1 mA tACS using the same gain settings, and compute the recovered EEG correlation after removing the stimulation artifact. If the recovered correlation is substantially below the EEG-only 93.5% value, the paper must qualify the 'concurrent EEG–tES recording' claim to state that the system records unsaturated artifact-contaminated waveforms rather than preserving microvolt EEG fidelity during stimulation.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that the platform can simultaneously record 8-channel EEG and deliver tDCS/tACS/tTIS without amplifier saturation. This claim depends entirely on the amplifier configuration used during the concurrent phantom recordings. Section II-A-2 states the front-end achieves a ±375 mV dynamic range, while Section III-C-2 reports that stimulation artifacts 'reach the volt range' and gives the example that 2 mA across ~5 kΩ corresponds to about 10 V. The concurrent trials used 1 mA, which would still produce roughly 5 V across the same load. A ±375 mV input window cannot accommodate such signals, so some lower-gain or attenuating configuration must have been used; however, the paper never reports the ADS1299 PGA gain, reference voltage, or input scaling for Figure 9. If gain was lowered to fit the artifacts, the LSB size and input-referred noise scale up, and the microvolt-level EEG fidelity claimed in the abstract and Section I is not established for the concurrent case. The raw plots demonstrate unclipped artifact recording, but they do not show that the underlying EEG signal is preserved. Because the novelty of the system is simultaneous high-rate EEG and tES, this missing configuration is the most load-bearing gap in the argument.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript presents a compact, MCU-centered bidirectional EEG-tES platform: an STM32 microcontroller with an ADS1299-based 8-channel EEG front-end sampling up to 8 kHz per channel and a dual-channel DAC/DDS constant-current stimulator supporting tDCS, tACS, tTIS, and programmable ramping. The authors report cable-based EEG correlation of 99%, gelatine-phantom EEG correlation of 93.5%, stimulation current amplitude errors below 1% across all tested modes, raw concurrent EEG+tES recordings without observed saturation, and computational performance metrics including latency, CPU load, power, and synchronization offset. The paper positions the system as a portable, low-complexity alternative to benchtop or FPGA-based integrated neuromodulation platforms. The main open questions concern the analog configuration used during concurrent recording, the methodology behind the phantom correlation figure, and the metrological support for the sub-1% current-error claims.","tokens_in":16012,"tokens_out":11670,"duration_ms":128668,"significance":"If the claims hold, this is a useful engineering contribution: it demonstrates that a mainstream MCU plus ADS1299 front-end and DAC-based current sources can sustain 8 kHz multichannel acquisition and generate tDCS/tACS/tTIS waveforms with small current error, which is relevant to portable closed-loop neuromodulation. The current-source characterization is well defined by Eqs. (1)-(2), and the use of an injected EEG template for fidelity assessment is a sound approach in principle. The weaknesses identified below are concentrated in the concurrent-recording claim, which is the paper's differentiating feature, and in the quantitative interpretation of the reported correlation and error values. These are addressable with additional reporting and targeted experiments, so the work is worth revising rather than rejecting.","major_comments":[{"comment":"The central claim of simultaneous EEG-tES recording 'without saturation' is not reproducible from the reported configuration. Section II-A-2 states that the front end achieves a ±375 mV dynamic range, while Section III-C-2 reports that the stimulation artifacts 'reach the volt range' and that 2 mA across approximately 5 kΩ corresponds to about 10 V; the concurrent trials used 1 mA, which would be roughly 5 V across a comparable load. These statements are inconsistent unless the ADS1299 PGA gain, reference voltage, and any input attenuation or DC-blocking path used during the Figure 9 recordings are disclosed. If a lower gain or attenuator was used to fit the artifacts, the input-referred LSB size and noise scale up, so preservation of microvolt-level EEG during stimulation is not established. Please specify the exact analog configuration and provide calibrated amplitude scales for Figure 9, or run a controlled recovery experiment with a known EEG template during tDCS, tACS, and tTIS to support the simultaneous-sensing claim.","section":"II-A-2 / III-C-2"},{"comment":"The 93.5% phantom EEG correlation is computed by sliding the template across the recorded signal and reporting the segment with maximal overlap. As written, this procedure selects the most favorable alignment after the fact and can inflate the reported coefficient, especially if the template contains strong alpha bursts or the search range is wide. Please report the window length, the search range, the selection rule, and whether the alignment was fixed before the analysis as a known-delay correction. In addition, report the correlation over the full recording or over pre-specified non-overlapping windows. Without these details, the phantom fidelity figure is not a well-defined performance metric.","section":"II-C-1 / III-C-1"},{"comment":"The sub-1% current-error claims lack metrological support. The current is computed from an oscilloscope voltage measurement across a 5 kΩ load, but the paper does not report the tolerance of the load resistor, the calibration of the oscilloscope, or an uncertainty analysis. Reported errors as small as 0.06% (Table I, tDCS 0.5 mA) are below the typical absolute vertical accuracy of the MSO44B unless additional calibration or averaging was performed. Please describe the calibration procedure, the resistor specification, and the measurement uncertainty propagation, or report the current errors as upper bounds consistent with the measurement setup.","section":"II-B-2 / Table I"},{"comment":"The concurrent-recording experiment demonstrates only that the ADC stream did not visibly clip and that stimulation-frequency components appear in the FFT; it does not demonstrate that the underlying EEG activity is acquired with usable fidelity. The text acknowledges that the EEG is 'strongly dominated' by stimulation artifacts, but the introduction states that the system should 'preserve microvolt-level neural activity while tolerating volt-level stimulation artifacts.' Please add a quantitative EEG-recovery evaluation during active stimulation, or explicitly limit the claim to unclipped broadband artifact capture and state that EEG recovery during stimulation is future work. This distinction affects the paper's core positioning, not just a secondary metric.","section":"III-C-2 / Section I"}],"minor_comments":[{"comment":"The tTIS caption states that reported values represent the mean error of the two stimulation channels, while the footnote says they represent the maximum absolute error of each channel; please make this consistent.","section":"Table I"},{"comment":"The middle row of Fig. 6 is described as the 'resulting amplitude envelope by linearly adding 2 tTIS channel outputs.' Please clarify in the text that this is an illustrative software sum of the two generated channels, not a measured electric-field interference pattern in tissue.","section":"III-B-3 / Fig. 6"},{"comment":"Figure 9 lacks calibrated amplitude scales, which is essential for assessing the saturation claim, and the caption uses '8 KHz' instead of '8 kHz.'","section":"Fig. 9"},{"comment":"The formatting of Table II is difficult to read because several columns combine numbers and units without clear separation, such as '1.67 34% 139.3'; please use a cleaner table layout.","section":"Table II"},{"comment":"The duration of the 'soft ramping' applied during all concurrent stimulation conditions is not stated; please report it, since ramping affects the transient artifact magnitude and the interpretation of the raw traces.","section":"Section III-C-2"}],"recommendation":"major_revision","confidential_remarks":"The reader's stress-test concern about the gain configuration lands: Section II-A-2 and Section III-C-2 are in tension unless the ADS1299 settings are disclosed. I recommend major revision rather than rejection because the missing configuration, the correlation methodology, and the absence of a quantitative concurrent EEG-recovery metric are all fixable. The concern about circularity is not applicable here; using the injected template as the reference is the correct fidelity-check approach. If the authors cannot supply a configuration consistent with the reported dynamic range, the 'without saturation / simultaneous EEG sensing' claim should be substantially weakened."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis paper is worth a look. The genuinely new thing is an MCU-centered platform that samples 8-channel EEG at 8 kHz while generating tDCS, tACS, and tTIS from the same STM32. That combination—high-rate recording plus kilohertz carriers in a wearable form factor—is not in the cited prior work, and if it works as described it fills a real gap. The benchtop stimulation numbers are believable: current errors below 1% across all modes on a 5 kΩ load, clean spectral peaks, and a sensible dual-buffer DDS design. The CPU and power budget table is also useful.\n\nThe soft spots are real but fixable. First, the 'without saturation' claim for concurrent recording is under-specified. The paper states a ±375 mV dynamic range for the ADS1299 front-end, yet Section III-C-2 reports stimulation artifacts in the volt range (1 mA across ~5 kΩ would be ~5 V). For the concurrent trials to fit, the PGA gain or some attenuation must have been changed, but the paper never reports that configuration for Figure 9. Without it, the no-saturation result is not reproducible, and the fidelity of the underlying EEG during stimulation is not established. The stress-test note gets this right.\n\nSecond, the 93.5% phantom correlation is computed by sliding the template across the recorded signal and reporting the maximal overlap. That's a best-case statistic. It should be disclosed as such or replaced with a fixed alignment. Minor, but it inflates the headline number.\n\nThird, the quantitative results appear to be single measurements with no error bars or repeats. For a hardware validation paper, that's thin, especially because device-to-device variation is the main risk in this kind of platform.\n\nFourth, no schematics, firmware, or data are released. For a paper whose contribution is a hardware platform, that is a significant reproducibility gap.\n\nNone of this sinks the core claim. The architecture is plausible and the benchtop measurements are sound. But the concurrent-recording evidence needs configuration details, and the statistics need more rigor. I'd send this to review—an expert referee can push for those fixes. It's the kind of paper that will be useful to engineers building closed-loop neuromodulation systems, and it's an honest attempt at an integrated wearable the field is missing.","headline":"A plausible MCU-based EEG-tTIS platform that fills a real gap, but the concurrent no-saturation claim needs configuration details before I'd trust it.","tokens_in":16672,"tokens_out":2968,"would_cite":true,"duration_ms":27882,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A compact wearable platform claims to record EEG at up to 8 kHz per channel while simultaneously delivering tDCS, tACS, and tTIS stimulation with current errors below 1%, and to capture both signals at once without amplifier saturation.","keywords":["EEG-tES integration","temporal interference stimulation","wearable neuromodulation","microcontroller architecture","direct digital synthesis","concurrent EEG and stimulation recording","closed-loop neuromodulation","8 kHz EEG sampling"],"falsifier":"Run the concurrent EEG–tES phantom test with a known microvolt-scale EEG template while delivering 2 mA tTIS at 2000/2040 Hz and check the raw ADC samples for clipping against the stated ±375 mV input range; if the volt-level artifact does not saturate, verify the front-end gain configuration used in that test and measure whether the microvolt EEG component can still be recovered from the mixture at the fidelity reported for EEG-only recording.","tokens_in":15611,"feed_emoji":"⚡","tokens_out":6976,"duration_ms":62003,"temperature":0.7,"pith_summary":"This paper tries to show that a small, battery-powered device built around one microcontroller can do what previously required benchtop or FPGA hardware: record eight channels of EEG at up to 8 kHz while simultaneously generating programmable tDCS, tACS, and temporal interference (tTIS) stimulation. The authors report current-amplitude errors below 1% on resistive loads, EEG waveform correlation of 99% in cable tests and 93.5% on a gelatine head phantom, and concurrent EEG–tES recordings that do not saturate or clip even when stimulation artifacts reach the volt range. If these results hold, the work would lower the hardware barrier for closed-loop neuromodulation—stimulation adjusted in real time from ongoing brain signals—and make kilohertz tTIS carriers directly observable in the recorded data rather than invisible to lower-rate EEG systems.","feed_headline":"One microcontroller records EEG and drives three stimulation modes","feed_subtitle":"A compact wearable device samples EEG at 8 kHz while delivering tDCS, tACS, and tTIS with under 1% current error.","key_machinery":"The load-bearing object is the single-microcontroller architecture with dual-buffer direct digital synthesis (DDS): while one waveform buffer streams to the DAC through DMA, the MCU computes the next buffer, so waveform output is continuous and programmable without a dedicated synthesizer or FPGA. On the sensing side, the front-end's configurable dynamic range (stated as ±375 mV) and high 8 kHz sample rate let the same data stream carry EEG and stimulation artifacts. The central quantitative identity is the measured current-error equation comparing programmed versus measured current amplitude; the architecture is what keeps that error below 1% while leaving CPU headroom (under 53% at 8 kHz with concurrent tTIS). Precision Howland current sources convert the DAC voltage to constant current up to ±4 mA into a 5 kΩ load.","core_discovery":"The central claim is that an MCU-centered architecture can integrate high-rate sensing and multimodal stimulation without FPGA-style resources. The platform combines a biopotential front-end sampled synchronously at up to 8 kHz per channel with a dual-channel constant-current stimulator built from a DAC plus microcontroller-based direct digital synthesis. On a 5 kΩ load, delivered current matches the programmed value within 1% for tDCS, tACS, and tTIS (carrier pairs such as 2000/2005 Hz), with negligible frequency error. Recording-validation experiments show 99% correlation with an injected EEG template over a cable and 93.5% through a gelatine head phantom; during simultaneous stimulation the raw traces contain the expected DC offset, sinusoidal artifact, or kilohertz carrier envelope without clipping. The paper concludes that this combination supports wearable, portable closed-loop neuromodulation and direct observation of stimulation waveforms during recording.","pith_inferences":["Beyond the paper: because CPU usage stays below 53% even at 8 kHz with concurrent tTIS, there is likely enough headroom to run artifact-removal or stimulation-parameter-optimization algorithms on the device itself; the paper does not implement such adaptive control.","Beyond the paper: the ability to record carrier signals directly could lead to a calibration step in which the system verifies the actual tTIS envelope delivered at the electrodes before each session, something the paper does not propose.","Beyond the paper: if the dynamic-range caveat is resolved, the architecture might scale to more sensing or stimulation channels to support multi-site temporal interference, though the paper only demonstrates eight EEG and two stimulation channels."],"forward_implications":["If the results hold, the same hardware can serve as a testbed for closed-loop algorithms that adapt stimulation parameters based on EEG in real time; the reported sub-millisecond synchronization offset supports tightly coupled triggers.","The 8 kHz sampling makes kilohertz tTIS carriers visible in recordings, enabling direct verification of delivered stimulation, artifact modeling, and possible dose control.","A battery-powered, MCU-based platform could move closed-loop neuromodulation out of the lab into home or point-of-care settings at lower cost than FPGA or benchtop systems.","The dual-buffer DDS supports non-periodic adaptive waveforms, so future protocols could use dynamically varying tACS or tTIS rather than fixed templates.","Measured current errors below 1% across 0.5–4 mA and both low-frequency and kilohertz bands suggest the output accuracy is sufficient for standard research protocols."],"supporting_citations":[{"why":"Defines tTIS as two kilohertz carriers whose interference envelope reaches deep targets; this is the stimulus the platform must generate and record.","marker":"[9]"},{"why":"Prior MCU-based tTIS stimulator that supplies the error-calculation equations and a comparison point for stimulation accuracy.","marker":"[22]"},{"why":"Provides the concurrent EEG–tACS artifact recording methodology and montage considerations used in the phantom concurrent experiment.","marker":"[28]"},{"why":"Supplies the gelatine head phantom formulation used to emulate the electrode–tissue interface.","marker":"[29]"},{"why":"FPGA-based closed-loop neuromodulation platform positioned as the power-intensive architecture the MCU design aims to replace.","marker":"[21]"},{"why":"Safety guidelines fixing recommended tES current limits, which motivate the 4 mA design ceiling.","marker":"[25]"},{"why":"Commercial benchtop closed-loop platform with tTIS support but a large footprint; baseline for portability comparison.","marker":"[18]"},{"why":"Commercial wearable EEG–tES device that lacks kilohertz tTIS; baseline for modality comparison.","marker":"[19]"}],"fun_headline_variants":["Single MCU records EEG and runs tDCS, tACS, tTIS","Wearable EEG-tES: 8 kHz recording, tDCS/tACS/tTIS","MCU-powered wearable does EEG and 3 brain-stim modes","EEG and temporal interference stimulation from one wearable","Wearable EEG-tES device: 8 kHz sampling, <1% current error"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the front-end was configured with enough dynamic range and low-enough gain during concurrent recordings to capture volt-level stimulation artifacts without saturation; the paper states a ±375 mV dynamic range but does not report the gain setting used in the simultaneous EEG–tES experiment, so the microvolt-level EEG fidelity during stimulation is not established.","fun_headline_variants_meta":{"raw":{"variants":["Single MCU records EEG and runs tDCS, tACS, tTIS","Wearable EEG-tES: 8 kHz recording, tDCS/tACS/tTIS","MCU-powered wearable does EEG and 3 brain-stim modes","EEG and temporal interference stimulation from one wearable","Wearable EEG-tES device: 8 kHz sampling, <1% current error"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001225,"raw_usage":{"total_tokens":5076,"prompt_tokens":1023,"completion_tokens":4053,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":639,"completion_tokens_details":{"reasoning_tokens":3961}},"tokens_in":639,"tokens_out":4053,"duration_ms":28243,"temperature":1.0,"reasoning_tokens":3961,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T20:49:12.669298+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the concurrent EEG–tES phantom test with a known microvolt-scale EEG template while delivering 2 mA tTIS at 2000/2040 Hz and check the raw ADC samples for clipping against the stated ±375 mV input range; if the volt-level artifact does not saturate, verify the front-end gain configuration used in that test and measure whether the microvolt EEG component can still be recovered from the mixture at the fidelity reported for EEG-only recording.","supporting_citations":[{"cited_title":"Noninvasive deep brain stimulation via temporally interfering electric fields,","cited_arxiv_id":null,"evidence_quote":"Defines tTIS as two kilohertz carriers whose interference envelope reaches deep targets; this is the stimulus the platform must generate and record."},{"cited_title":"Designing and pilot testing a novel tran- scranial temporal interference stimulation device for neu- romodulation,","cited_arxiv_id":null,"evidence_quote":"Prior MCU-based tTIS stimulator that supplies the error-calculation equations and a comparison point for stimulation accuracy."},{"cited_title":"Removal of gross artifacts of transcranial alternating current stimulation in simultaneous eeg monitoring,","cited_arxiv_id":null,"evidence_quote":"Provides the concurrent EEG–tACS artifact recording methodology and montage considerations used in the phantom concurrent experiment."},{"cited_title":"Investigating gelatine based head phantoms for electroencephalography compared to elec- trical and ex vivo porcine skin models,","cited_arxiv_id":null,"evidence_quote":"Supplies the gelatine head phantom formulation used to emulate the electrode–tissue interface."},{"cited_title":"Low intensity transcranial electric stimula- tion: Safety, ethical, legal regulatory and application guide- lines,","cited_arxiv_id":null,"evidence_quote":"Safety guidelines fixing recommended tES current limits, which motivate the 4 mA design ceiling."},{"cited_title":"Neuroconn loop-it","cited_arxiv_id":null,"evidence_quote":"Commercial benchtop closed-loop platform with tTIS support but a large footprint; baseline for portability comparison."},{"cited_title":"Starstim tes-eeg systems","cited_arxiv_id":null,"evidence_quote":"Commercial wearable EEG–tES device that lacks kilohertz tTIS; baseline for modality comparison."}],"review_version":1}