Exploring the limitations of event-related potential measures in moving subjects. Case studies of four different technical modifications in ergometer rowing

Measuring brain activity outside the laboratory is of great importance for investigating human behavior under naturalistic conditions (e.g., in cognition and movement research, application of brain-computer interfaces). To measure neuronal activity in moving subjects, only modified NIRS and EEG systems are applicable. Because conventional EEG systems are too sensitive to movement artifacts, artifact sources should be eliminated beforehand to improve signal quality. Four different approaches for EEG/ERP measures with moving subjects were tested in case studies: (i) a purpose-built head-mounted preamplifier, (ii) a laboratory system with active electrodes, (iii)+(iv) a wireless headset combined with (iii) passive or (iv) active electrodes. A standard visual oddball task was applied during rest (without movement) and during ergometer rowing. All 14 measures revealed very similiar (within subjects) visual evoked potentials for rowing and rest. The small intraindividual differences between rowing and rest, in comparison to the typically larger interindividual differences in the ERP waveforms revealed that ERPs can be measured reliably even in an athletic movement like rowing. The expected modulation of the motor-related activity by force output, on the other hand, was largely affected by movement artifacts. Therefore, for a successful application of ERP measures in movement research, further developments to differentiate between movement-related neuronal activity and movement-related artifacts are required. However, it cannot be excluded that activities with small magnitudes related to motor learning and motor control cannot be detected because they are superimposed by the very large motor potential which increases with force output.


Introduction
The investigation of brain functions with noninvasive methods like functional magnetic 3 resonance imaging (fMRI), magnetencephalography (MEG), electroencephalography (EEG), 4 near-infrared spectroscopy (NIRS), and transcranial magnetic stimulation (TMS) is typically 5 limited to laboratory settings. This is because the systems are very large and cannot be moved 6 and head movements must be avoided (fMRI, MEG) or head and body movements generate 7 large movement artifacts. Within the last decade there has been a rapidly increasing interest in 8 investigating brain functions in ecological settings (e.g., cognitive or neuropsychological 9 processes in interaction with a natural environment or in a social context; and movement 10 analysis and motor learning) which require portable systems (e.g., Ruffini 12 Especially in the field of movement research/motor learning laboratory settings are strongly 13 limiting because only simple finger or hand or arm movements can be investigated and it is 14 questioned whether the results of these studies can be transferred to complex movements (e.g. 15 Dum and Strick, 2002; Hazeltine and Ivry, 2002;Wulf and Shea, 2002). In principle, NIRS 16 and EEG are suited for measures in moving subjects because the sensors are small and fixed to 17 the head (rather than the head being fixed to the sensor) and the necessary electronics and 18 recording devices can be built small enough. NIRS, like fMRI, measures the hemodynamic 19 response related to specific brain processes with a low temporal resolution. Spatially it is 20 restricted to cortical layers close to the skull. Atsumori 25 showed that the task was performed successfully and comparably in all three conditions, 26 whereas data loss was highest in real cycling (about 35%), but much lower in indoor cycling 27 (7.5%) and under rest (5%). 28 Event-related potential (ERP) measures, which are the focus of this paper, have the 29 well-known limited spatial resolution but a high temporal resolution which is essential to 30 analyze complex movements: for example the time course of feedback and feedforward 31 processing in visuomotor learning (e.g. Hill, 2009Hill, , 2014. Conventional EEG systems are very 32 sensitive to mechanical (cable and electrode movements) and physiological (electromyogram 33 (EMG) of head and neck muscles, and sweating) movement artifacts (Zschocke, 2002). If 34 cognitive processes in a moving subject rather than the movement itself are the focus of 35 interest, data preprocessing algorithms informed by the behavioral movement data can be used 36 to clean the EEG data from movement-related (neuronal and artifactual) activity (Gwin et al.,37 2010; Stone et al., 2018). If the motor-related activity is of interest, advanced data 38 preprocessing algorithms such as independent component analysis (ICA; Makeig et al., 1996) 39 can be used to correct such artifacts. However, a study measuring ERPs during walking and 40 running on a treadmill showed that data loss was very high (on average 130 of 248 EEG channel 41 signals remained), even using a system with active electrodes which are considerably less prone 42 to artifacts than conventional passive electrodes Gwin et al., 2011). 43 Furthermore, the usability of conventional EEG systems like the one used in these studies is 44 limited for measures with moving subjects. In movement tasks with only marginal head 45 movements, like cycling on an ergometer, these laboratory systems combined with ICA-based 46 artifact correction can be applied succesful for EEG measures ). E.g. 47   found in a high-intensive cycling exercise an increase in spectral power 48 when the athletes were fatigued. If spectral changes of higher EEG frequencies (alpha to 49 gamma) are in the focus of interest, like in the Enders et al. study, artifacts directly coupled to 1 movement execution are outside of this frequency range because movement frequencies are 2 considerably lower. However harmonics of these movement frequencies may occur which have 3 to be considered. With fully moving subjects, in contrast, artifacts are more difficult to handle. 4 In a cocktail party study (including eating, drinking, chatting, etc.) with ten subjects wearing 5 self-made (noncommercial) wireless EEG headsets, about 40% of the data was lost due to 6 artifacts in contrast to 4% in two laboratory studies (Gevins et al., 2012). Despite this high data 7 loss, these studies revealed valuable ERP Gwin et al., 2011) or spectral 8 EEG (Gevins et al., 2012) results. However, especially in movement research the method of 9 choice is to avoid the generation of mechanical artifacts beforehand by technical modifications. 10 This approach was used in the four pilot studies reported in this paper that tested the suitability 11 of different technical solutions to measure ERPs during ergometer rowing (Fig. 1). Especially 12 for analysing motor-related brain activity with ERPs, rowing is well suited. It is a cyclic 13 movement with a high number of repetitions and the degrees of freedom of the movement are 14 limited by the biomechanical constraints of the equipment (boat and scull/oar, or ergometer). 15 Furthermore, in contrast to e.g. cycling or cayaking, the rowing movement is composed of 16 different (more or less) distinct movement elements and movement frequency is lower (20-40 17 strokes/minute vs. 60-120 revolutions/minute in cycling). Finally, the biomechanical data 18 (dynamics and kinematics of the movement, and boat and scull/oar movement), which are 19 partly necessary for ERP analysis, can be measured meanwhile with relative ease. 20 The most critical question before pilot Study 1 (see below) was, if movement artifacts 21 distort the EEG data completely or if there are systematic artifacts which can be controlled and 22 either (partly) excluded or corrected in offline analysis. To test this question, a reliability check 23 was made. A standard visual oddball task was applied in a rest condition (without movement) 24 and during ergometer rowing. Similarnot movement related -ERP activities during rest and 25 during rowing would show that ERPs can be measured reliably in a moving subject. The second 26 methodological questionconcerning the analysis of motor behavior with ERPsis: can 27 movement-related artifacts be identified and separated from motor-related neuronal activity? 28 As an indicator for a reliable measure of motor-related activity, at least a motor potential (MP), 29 a negative activity related to force output, should be expected. Siemionow et al. (2000) showed 30 in a study which used isometric elbow flexions that the amplitude of the MP (labelled motor-31 related cortical potential (MRCP) in this study) correlates very highly with EMG activity (r > 32 0.8) and the generated muscle force (r = 0.95). Furthermore, Dai et al. (2001) found in a fMRI 33 study a high correlation between isometric force (using a hand-grip dynamometer) and activity 34 in the primary motor cortex. The MP/MRCP must be generated by different generators because 35 it can already be evoked when a muscle activation is only imagined. This should originate in 36 the supplementary motor area (SMA; Ranganathan et al., 2004). A second part of the MP is 37 thought to be related to the control of muscle activation by the primary motor cortex.  45 small, purpose-built 20-channel system was used with a preamplifier connected to the head and 46 electrodes with shortened and fixed cables mounted to a standard electrode cap. Because this 47 approach was only partly successful the second approach (Study 2) was performed. This 48 experiment involved the use of a to that time (2008) in our lab available system with active 49 electrodes with built-in preamplifiers, and the amplifier was worn in a backpack. In pilot Study 1 3 (2013), as the newest improvement in EEG technology, a small head-mounted EEG system 2 with wireless data transmission was combined with the electrodes and cap used in Study 1. In 3 Study 4 (2018), finally, the headset used in Study 3 was combined with active electrodes and 4 rowing force and movement speed was systematically varied to investigate the influence of 5 these factors on data quality. used ergometers for training in competitive rowing and performance diagnostics. In the drive 12 phase of the rowing cycle, the rower's pull accelerates an air resistance braked flywheel in the 13 round cage. In the recovery phase the rower moves in opposite direction on the sliding seat, 14 preparing for the next pull. A monitor displays stroke rate, time, and distance rowed, power per 15 stroke, mean power, calories burned. On the chair at the right side in front of the rower is the 16 laptop for stimulus presentation. The laptop on the table recorded the behavioral and EEG data 17 (photo use with participant's permission).

23
A preliminary test using a standard EEG system (NeuroScan SynAmps) by wearing 24 the 32-channel preamplifier headbox in a backpack revealed no satisfying results. Therefore, 25 to reduce the generation of artifacts due to cable and electrode movements, a small purpose-26 built (according to Simon, 1977) battery-powered 20-channel system (based on positive 27 previous experiences with a similar three-channel system) was used with an occipitally 28 mounted preamplifier (differential amplifier, gain = 30; hardware filters: 0.27 Hz passive RC 29 highpass; 30 Hz 2nd-order Bessel lowpass). After a second amplifier stage (total gain = 3600) 30 6 the signals were digitised using a BEST system (Dr. Grossegger & Drbal, Korneuburg, Austria; 1 sample rate 256 Hz, resolution 0.3 μV/bit). An electrode cap (EasyCap, www.easycap.de) and 2 22 gold electrodes (Grass) with shortened cables were used. Electrode caps with electrodes 3 fixed to the cap have the disadvantage that the contact of some electrodes with the skin can be 4 poor, depending on head shape (that is, if there are dents in the skull the distance between cap 5 and skin can be too large). Therefore the standard adaptors inserted in the cap were removed, 6 the adaptor holes were enlarged, and the electrodes were fixed using a highly viscous 7 conductive paste (Nihon-Kohden Elefix). EEG was recorded from 20 sites (midline: FPz, Fz, 8 FCz, CPz, Pz, and Iz; Left/right: FC1/2, FC3/4, FC5/6, C3/4, CP1/2, CP3/4, left and right 9 mastoid) covering mainly the sensorimotor area, and Iz to assess visual evoked potentials 10 (VEPs). Data were recorded using Cz as reference and rereferenced offline (see results for 11 details). A visual oddball task was used with 180 black-white fullscreen checkerboards 12 (frequent), 60 red-white checkerboards (deviant), and 60 gray crosses (target) which 13 had to be counted. Stimulus duration was 100 ms, the interstimulus interval varied randomly 14 between 800 and 1000 ms. The oddball task is a standard paradigm and generates robust ERP 15 components: VEPs at occipital sites which are generated in the visual cortex when visual 16 stimuli are presented. The P300 at centroparietal sites (around CPz, Pz) when attention is 17 shifted to a target stimulus. Stimuli were presented asynchronous to the temporal pattern of the 18 rowing movement on a computer screen besides the ergometer using the Presentation software 19 (Neurobehavioral Systems, www.neurobs.com), allowing to monitor the display without head 20 movements, although eye movement artifacts may be generated. Alternatively, an acoustical 21 stimulation could be used. However, because the auditory cortex is closer to the sensorimotor 22 cortex and to the mastoids than the visual cortex, this will probably lead to extended signal 23 overlay of motor-related and auditory-evoked potentials. The EEG was analyzed using the  Fig. 2 displays ERPs of all three subjects for the oddball task. Especially the VEPs were 4 very similar between rowing and rest, and intraindividual differences were much smaller than 5 the interindividual differences. This result demonstrates that standard ERPs not time-locked to 6 the rowing movement can be measured during rowing, despite the fact that subjects M and J 7 had no or only marginal rowing experience. In contrast, the motor-related ERPs were very 8 noisy and showed large artifacts with large inter-and intraindividual (implemented by varying 9 force output and stroke rate) differences and were therefore not interpretable. One observable 10 source of large artifacts was due to movements of the cable connecting the head-mounted 11 preamplifier with the second amplifier unit, which led to movements of the preamplifier and 12 electrode cables. waveforms. The high dropout rate for subject H was mainly due to eyeblink artifacts. For 25 subject M three channels (FCz, CPz, and CP4) were lost completely due to artifacts during 26 rowing: for subject J the left mastoid channel was lost. Furthermore, a P300 is missing during 27 rest for J because he did not count the target because of an imprecisely given instruction. For 28 M and H the P300 was larger during rest. The ERPs of the deviant stimulus condition are not 29 displayed because they showed results similar to those of the frequent condition.

4
The second approach used a system with active electrodes at the Department of 5 Psychology, University of Frankfurt. This system suppresses artifacts due to cable movements, 6 however electrode movements cannot be avoided completely. During rowing and rest the 7 amplifiers were worn in a backpack and connected via fibre-optic cables to a PC. EEG was 8 recorded continuously with BrainAmp DC amplifiers (BrainProducts, Gilching, Germany; 9 sample rate 250 Hz, resolution 0.1 μV/bit, input impedance 10 MOhm) using an equidistant 10 EasyCap (EasyCap GmbH, Herrsching-Breitbrunn, Germany, www.easycap.de) with 62 11 sintered Ag/AgCl electrodes and built-in preamplifiers (BrainProducts ActiCap System). Eye 12 blinks and movements were monitored with supra-and infra-orbital electrodes and with 13 electrodes on the external canthi. The vertex electrode was used as the reference. To avoid 14 injuries due to skin abrasion, electrode impedances were kept at 20 kOhm which is more than 15 sufficient from electrical engineering principles (Ferree et al., 2001; see Picton et al., 2000). 16 The EEG was analyzed like in Study 1 with slightly different filter settings (0.5 to 20 Hz  Fig. 3 displays ERPs of all three subjects for the oddball task. As in Study 1, the VEPs 28 were very similar between rowing and rest, the P300 was smaller during rowing, and the 29 intraindividual differences were much smaller than the interindividual differences. 30 Motor-related activity, on the other hand, was not dominated by the expected motor 31 potential in all conditions. Applying an ICA-based artifact rejection procedure improved the 32 data quality somewhat but insufficiently, probably because movement artifacts are not stable 33 enough over time, which is a prerequisite for computing the ICA components. Performing only 34 the arm pull revealed for subject H a bilateral negativity; however, for subjects F and K it 35 revealed no interpretable results. For normal rowing a negative activation at central sites, 36 indicating a motor potential, appeared, but at peripheral electrode sites large activities occurred 37 which were only partly corrected using ICA. One possible source for artifacts in this study 38 were electrode movements due to cable drag because cables were not fixed to the cap. 39 In summary, this pilot study revealed results similar to those of Study 1. Standard ERPs 40 can be measured reliably during rowing, however motor-related activity is largely distorted by 41 remaining artifact sources.  (14). In addition, the sensor locations cannot be changed and are not well suited to sensorimotor 28 research. Therefore, as already practised by Debener et al. (2012), this system was modified by 29 removing the original sensors and plastic arms and connecting the system to the electrode cap 30 with the gold electrodes used in Study 1 (Fig. 4). Comparison studies revealed that the Emotiv Signal quality was controlled with the Emotiv Testbench recording software. The ergometer, 8 its measuring equipment, and the oddball task (except a reduction of the number of standard 9 stimuli from 180 to 120) were the same as in Study 1. The synchronisation of the EEG data 10 with the biomechanical data and the visual stimulation was somewhat difficult. Although the 11 Emotiv system can read trigger signals from a serial port, the laptops used did not have serial 12 ports. An interface with serial-to-USB adaptors is not accurate enough in timing. Therefore, 13 the Presentation scenarios were modified for synchronisation. A photodiode was fixed to the 14 Presentation laptop and activated by stimuli at the beginning and the end of each experimental 15 run. The triggers sent via the parallel port were recorded with an USB analog-digital device 16 (RedLab 1208 LS, Meilhaus, Puchheim/Germany) together with the biomechanical data (force 17 of the rowing stroke and movement of the sliding seat). The signal of the photodiode (which 18 was much larger than the EEG) was recorded with one EEG channel, and the cables were 19 removed during the experimental runs. The EEG, biomechanical data, and triggers of the 20 oddball stimulation were synchronised offline using purpose-written software. A separate 21 channel for received data packets implemented in the Emotiv system provides the ability to 22 control for lost data. This infomation was used to correct the synchronisation (a small fraction 23 of samples was lost in 4 of 25 datasets). The EEG was analyzed in a way similar to that used 24 in Study 1 (using only a 20 Hz lowpass filter). Data of the rowing condition of two subjects 25 had a larger number of trials contaminated with eye blinks. These were corrected successfully 26 using ICA without affecting other activities (despite the low spatial resolution relying on only 27 14 channels can be critical to apply the ICA procedure).

40
Fig. 5 displays ERPs of all four subjects for the oddball task. As in studies 1 and 2, the 41 VEPs were very similar between rowing and rest, the P300 was smaller during rowing, and the 42 intraindividual differences were much smaller than the interindividual differences. 43 The motor-related activity showed more or less large artifacts at several sites, as well 44 as the expected modulation by rowing force at some other sites --that is a larger negative 45 activity with increasing force during the rowing stroke --with some differences between 46 subjects (Fig. 6). As it can be difficult to differentiate between motor-related activity and 47 movement-related artifacts at sensorimotor sites, movement-related artifacts can clearly be 48 identified at sites outside the sensorimotor region. Fig. 5 displays such an example for subject 49 J at electrode site O1. The VEP for the frequent stimulus of the oddball task was computed 1 separately for the drive and recovery phases of the rowing movement (stimuli presented in the 2 transition drive-recovery were excluded for this analysis). These waveforms were overlaid by 3 large movement-related artifacts with reversed polarity which cancelled each other out in the 4 average including all trials. These artifacts were independent of the chosen reference (Cz, right 5 mastoid, and linked mastoids). That is, the overlay does not depend on physiological motor 6 activity, which is much stronger at Cz than at the mastoids and O1; instead, it must be generated 7 at electrode O1. Probably this was due to cable artifacts in an electromagnetically noisy 8 environment. Furthermore, the electrodes O1, left mastoid, and AFz had the largest distance to 9 the Emotiv connector and therefore the longest cables, which may have allowed small 10 movements. 11 To identify one possible source of movement artifacts, the gyroscope of the Emotiv 12 system was used to test if artifacts are caused by head movements. Therefore, rapid repeated 13 movements were performed (left turn, right turn, and nodding) revealing very large artifacts, 14 especially at lateral sites where the impact of the head movement was larger than at central 15 sites. Using ICA these artefacts could be strongly attenuated (Fig. 7). A comparison of the 16 gyroscope data for these head rotations and rowing showed only small head movements for 17 nodding during rowing: that is, head rotations and the associated artifacts are not critical for 18 rowing. 19 In summary, this pilot Study 3 revealed results similar to those of Studies 1 and 2.  Under combining the technical advantages of Study 2 (active electrodes) and Study 3 14 (mobile headset), Study 4 followed two aims: (i) can movement artifacts be reduced when the 15 passive electrodes of Study 3 are replaced by active electrodes, and can movement-related 16 ERPs be measured in sufficient quality then? This would be the optimal result. (ii) If the first 17 aim is not achieved: how is the signal quality of standard VEPs affected by movement 18 dynamics (force output) and movement kinematics (movement speed). That is, is there a trade- 19 off between signal quality and movement intensity in the measurement of ERPs in e.g. 20 cognitive tasks in moving subjects.

24
The Emotiv system used in Study 3 was combined with eight active electrodes 25 (EasyCap active) which were provided by EasyCap GmbH (www.easycap.de) together with 26 electrode caps (EasyCap). In contrast to the older version of the Acticap electrodes used in 27 Study 2 these electrodes are smaller and very flat, and should therefore be less susceptible to 28 tilting movements generated by inertial forces. After pretests electrode cables were shortened, 29 fixed to the cap and connected to the headset. Power for the electrodes was provided by a 9 V 30 battery attached to the connector (Fig. 8). The electrodes were fixed with adaptors and a highly oddball task were the same as in Study 3 except two modifications. The condition with the 60 7 deviant stimuli of the previous studies was removed because the VEPs did not differ from the 8 standard checkerboard task. The number of trials of the latter was therefore increased from 120 9 to 180. Secondly, because from one subject no VEP signal was obtained, a second recording 10 session was conducted at a later time (with a nearly identical rowing performance). To ensure 11 that the visual stimulation was observed, the target stimulus (cross) was replaced by a lower 12 number (27-33) of pictures of different airplanes. This was done because the oddball task was 13 repeated six times instead of two times in the previous studies and therefore more salient stimuli 14 were used. In this Study 4 the target condition of the oddball task was only used to control 15 performance. 16 The same four male subjects as in Study 3 (aged 19, 23, 26, and 59 years, height 180- 17 190 cm, weight 80-84 kg) performed six experimental conditions of about four minutes 18 duration comparable to Study 3: (i) rowing with lower force output and lower stroke rate; (ii) 19 lower force output, higher stroke rate; (ii) higher force output, lower stroke rate; (iv) higher 20 force output, higher stroke rate; (v) visual stimulation in rest (without rowing); (vi) rowing 21 with arm pull only (details are provided in the legend of Fig. 10). Power and stroke rates could 22 be controlled with the performance monitor of the ergometer and were close to the instruction. 23 The rowing power ranged from recreational (condition i) to long distance (e.g. 10 km) racing 24 (condition iv). The visual stimulation was applied in all six conditions. For comparison, one 25 subject repeated all six conditions with the cap used in Study 3 with 14 passive electrodes.

7
Fig. 9a displays the VEPs generated by the standard stimulation (checkerboard) of the 8 oddball task for all four subjects and conditions. For direct comparison, the VEP measured at 9 active electrode O1 and passive electrode Oz are superimposed. As in the previous studies the 10 VEPs were similar between rowing and rest, and the intraindividual differences were much 11 smaller than the interindividual differences. Remarkably, the VEP measured at O1 and Oz were 12 highly congruent. Fig. 9b displays the Standard Deviations (SD) of the EEG segments from 13 which the average VEP was computed for all subjects and the six conditions. SDs were smallest 14 in condition five (rest) and largest in the high intensive rowing condition four (except subject 15 J). The signal-to-noise ratio (SNR) was computed in addition and revealed a similar pattern 16 than the SD, that is the highest SNR in condition five and the lowest in condition four. However 17 SD as well as SNR values could be misleading, if an average waveform is dominated by a large 18 and systematic artifact. Nevertheless these data show that the qualitiy of the ERP is lower when 19 movement intensity increases, as is indicated in the VEP waveforms as well. both methods reducing cable movement artifacts. A further advantage of meanwhile available 27 systems with low weight and small dimensions for movement research resp. research on 28 moving subjects is either wireless data transmission or storing the data on a SD-card in the 29 device itself. 30 The data from fourteen single case measures from six different subjects revealed for a 31 standard paradigm (visual oddball task) comparable intraindividual ERPs during rowing and 32 during rest (non-movement condition) in all cases, despite remaining artifacts in the data. EEG 33 parameters and ERP waveforms are genetically determined and show generally a broad range 34 of interindividual differences, but are also remarkably stable over time in adult subjects 1 (Stassen et al., 1998;Weisbrod et al., 2004). This fact (although probably not well known), 2 strongly supports the reliability of the data, because the intraindividual differences were much 3 smaller than the interindividual differences. Higher intraindividual differences instead would 4 indicate that the ERP pattern are largely distorted by movement artifacts. 5 Whereas the VEPs were quite similar, the P300 was smaller during rowing than during 6 rest (studies 1-3). This may be partly due to a habituation effect (the rest condition was always 7 performed before the rowing condition), like recently suggested by Scanlon et al. (2017) as 8 well, who used also an oddball task to compare ERP measures during indoor cycling with a 9 resting condition. Another contribution to this effect was probably that multiple task demands 10 reduce the P300 (e.g., as reviewed by Polich, 2007). Here, attention is divided by counting the 11 targets on the one hand and performing the rowing movement and the monitoring of stroke rate 12 and power output on the other hand, which may be demanding for nonskilled rowers. Similar 13 results with about a 30% smaller P300 amplitude during walking compared to sitting still (in also suggested that the different task demands were the reason for this result. However, it has 16 to be emphasised that the aim of the present pilot studies was not to investigate cognitive 17 processes in rowing, instead these robust ERP components (VEPs, P300) were measured for 18 methodological reasons, that is to compare ERP data quality during rowing and rest in a 19 repeated measure design. 20 The positive results of the oddball task are promising for the investigation of brain 21 functions in naturally behaving subjects outside the laboratory: for example in cognition 22 research, brain-computer interface (BCI) applications, ambulatory assessment, and others. 23 Since rowing is a very athletic sport and therefore a source of large movement-related artifacts, 24 ERP measures with less motor activity such as walking around (e.g., Debener et al., 2012) or 25 cycling when pedaling slowly at a subaerobic level (Scanlon et al., 2017) can easily be obtained 26 using suitable equipment. 27 Due to the nature of ecological settings more limitations compared to laboratory 28 settings have to be accepted. The placement of the stimulation monitor besides the ergometer 29 and the oscillating viewing distance may be regarded as critical. Alternatives might be the use 30 of a head-up display or an acoustical stimulation. However the gyroscope data in studies 3 and 31 4 revealed only small head movements in the oddball task during rowing, therefore it can be 32 concluded that this was comparable in Study 1 and 2 because the placement of the monitor was 33 the same. Eye movements were marginal or not present in the high densitiy recording of Study 34 2 but could be monitored at frontal sites in the other studies as well. 35 The (slight) differences in hardware and filter settings, and electrode positions were 36 acceptable because reliability was assessed by the within-subject comparisons for each study 37 separately. 38 The small sample size of these four pilot studies may be seen as critical from a classical 39 cognitive neuroscience point of view. However it has to be noted that no subtle cognitive 40 processes were investigated which require a larger sample size. Instead, robust 41 electrophysiological (i) and physical (ii) processes were investigated which were seen in all 42 fourteen measures. That is (i) VEPs were obtained during rowing and rest in all measures. 43 And (ii) it was not possible to measure clean movement related neuronal activities. That is the 44 chosen technical setup clearly differentiates between possibilities and limitations. 45 The second and more challenging aim of the four studies was to test if motor-related 46 activity could be measured. This approach is unique for these pilot studies as the cited studies 47 in the introduction aimed to measure cognitive processing in movement conditions, not the 48 movement itself. Although motor potentials during the drive phase of the rowing movement, 1 modulated by force output, were indicated (cf. Figure 6 and 10), in all four studies large 2 movement-related artifacts occurred which distorted motor-related activity. These artifacts can 3 be identified at electrode sites apart from the pre-and primary motor cortex. In this context it 4 has to be considered that artifacts originating from the reference electrode will affect the other 5 electrodes. As the classical mastoid reference captures EMG activity of head and neck muscles, 6 reference electrode positions less affected by this EMG activity, as well as the activity of 7 cortical motor areas, may be better suited (e.g. prefrontal sites or nose tip). At sites covering 8 the motor areas artifacts are more difficult to detect because muscle force generation, 9 movement kinematics, and movement-related artifacts have the same time course. Therefore, 10 further technical improvements to reduce artifacts beforehand or to identify artifacts better (cf. 11 Castermans et al., 2014) and correct them are necessary to investigate motor behavior in 12 movements including the whole body (as in sports: for example, motor learning or 13 differentiating high from low performance in movement execution). One example to identify 14 artifact sources was given in Study 3 when using the gyroscope to identify artifacts generated 15 by rapid head movements. 16 Known artifact sources are the EMG activity and sweating artifacts. The latter cannot 17 be filtered out when movement frequency is in the same range (like in one subject of Study 4). 18 Other sources of artifacts may rely on small movements of the electrode cables relative to the 19 cap which were still possible; and the translational head movement during rowing in an 20 electromagnetically noisy environment (the room was not shielded) may have generated small 21 currents in the cables, as in a generator (according to Faraday's law). However in both cases 22 the active electrodes should be less vulnerable to this artifact sources. Furthermore, for the 23 second case, artifacts should be larger when movement speed increases which was not 24 observed. 25 To investigate further possible sources of motion artifacts, Oliveira et al. (2016) used a 26 phantom head to simulate motion artifacts in EEG data and found that artifacts increased with 27 movement frequency as well as with movement amplitude, that is in general with the 28 acceleration of the phantom head. "We speculate that the major source of such artifacts is 29 micro-movement of the recording electrodes in relation to the scalp surface" (Oliveira et al., 30 2016). Their data showed that artifacts strongly increased when the head acceleration was 31 larger than 1.5 g. Based on these results, additional measures of head acceleration using a 32 triaxial acceleration sensor (Move II, Movisens GmbH, Karlsruhe/Germany, 33 www.movisens.com) attached to the Emotiv headset were analysed. These revealed values 34 between 0.85 g in low intensive rowing (75 W, 20 strokes/min) and 2.5 g in high intensive 35 rowing (360 W, 30 strokes/min). That is the lower quality of VEPs in Study 4 at the rowing 36 conditions with higher intensity may partly depend on such micro-movements of the electrodes, 37 independently if passive or active electrodes are used because the main advantage of active 38 electrodes is that these are less susceptible for cable sway artifacts. 39 Another pitfall with a physiological origin might be, that different neuronal activities  Study 4). In contrast to these, probably from the pyramidal cells in the primary motor 45 cortex generated efferent activities (Brecht et al., 2004;Shibasaki, 2012), are other motor-46 related activities (originating from the premotor cortex or the SMA) very small. E.g. in two 47 own visuomotor tracking studies the effects related to motor learning were below 1 μV (Hill, 48 1 (e.g. in rowing the perception and adaptation of within-crew differences of rowing technique, 2 Hill, 2002) are in this amplitude range, these will be hidden when the activity related to force 3 execution is much higher.