g.BSanalyze: Specs & Features
Perform offline biosignal analysis under MATLAB, or stand alone with powerful toolboxes
g.BSanalyze - gtec's Biosignal Analysis Software
g.BSanalyze is an interactive environment for multimodal biosignal data processing and analysis in the fields of clinical research and life sciences. g.BSanalyze has been on the market for more than ten years, and is used in more than 70 countries. It is the most comprehensive package to analyze non-invasive and invasive brain-, heart- and muscle-functions and dysfunctions. The package won several international Awards. The new version includes many new functions such as support vector machines, event-related ECG, support for P300 and SSVEP/SSSEP BCIs, zero class detection for BCIs, compressed spectral array, minimum energy, and more!
g.BSanalyze consists of a base version for data import, visualization, transformation and pre-processing and has several dedicated toolboxes:
- EEG toolbox: specialized functions for pre-processing, analysis and parameter extraction for EEG data
- ECG toolbox: find QRS complexes and calculate heart rate variability parameters
- Classify toolbox: classify parameters with linear and non-linear methods including statistical analysis for zero class detection
- High-resolution EEG: map EEG activity on realistic head models
- CFM toolbox: calculate amplitude integrated EEG
- SPIKE toolbox: analyze spikes, multi-unit activity and positions to map physiological parameters
The package comes with many sample biosignal data-sets, including P300, SSVEP, motor imagery, CSP BCIs, Tilt-Table, EPs, multi-unit activity, CFM, and ERD/ERS.
g.BSanalyze's graphical user interface includes more than 100 state-of-the art functions for defining electrode montages, spatial or temporal filter designs, artifact treatment, quality control, spectral analysis, coherence, correlation, bandpower analysis, ERD/ERS analyses, EP analyses, visualization, data set classification, and other goals. It is the only package that supports all BCI principles: P300, SSVEP/SSSEP, motor imagery and slow cortical potentials. You can load and save your preferred processing steps within a script program and automatically process your data in g.BSanalyze batch mode.
g.BSanalyze's processing capabilities allow you to extract relevant features from your multimodal data and define useful parameters for postprocessing. Use these parameters directly with g.BSanalyze's classification tools to assign distinct classes to your data with linear and non-linear classifiers.
The combination of the graphical user interface and the programming environment makes g.BSanalyze a truly unique package for biosignal analyses.
The stand-alone version of g.BSanalyze can run without a Matlab installation, but batch processing in the Matlab command window is not possible.
- interactive and intuitive graphical user interface for EEG, ECoG, EOG, EMG, ECG, spikes, and physical data analyses and documentation under MATLAB as well as a stand-alone version
- extensive tools for data processing in time, space, and frequency domains
- powerful 2-D and 3-D visualization tools to rapidly generate publication ready figures
- enhancement of power with g.tec's specialized EEG, aEEG, ECG, SPIKE, CLASSIFY and High-Resolution EEG toolboxes
- flexibility to integrate other MATLAB toolboxes, as well as customers' specific algorithms
- analyze data from: g.Recorder, Highspeed On-line Processing for Simulink, MATLAB and C API and many other 3rd-party recording devices
- more than 100 state-of-the-art functions for analyzing biosignal data
- 15 years of development and used in more than 70 countries worldwide
- the only package that supports all BCI principles: P300, motor imagery, SSVEP/SSSEP, slow cortical potentials
- optimized for high-gamma activity analysis
The Base Version of g.BSanalyze allows the visualization, processing and basic analyses of EEG, ECoG, ECG, EOG, EMG, spikes, respiration, pulse, and other physical signals. An intuitive data editor lets you scroll through the data set, adding annotations and comments. Semi-automatic artifact detections and manual correction possibilities yield the highest quality data for further analysis. Data set triggering on events and event-related signal changes can be performed based on markers and signal channels. Temporal filtering and spatial filtering (e.g. Common Spatial Patterns, ICA, PCA) allow extracting hidden information from data sets.
The g.BSanalyze data editor lets you scroll comfortably through your data set. You have the choice between manual stepping, or taking advantage of the data player, which allows for automatic stepping at a user defined speed.
You can add/remove comments to specific data segments or mark channels or whole trials using different attributes (e.g. using marker ARTIFACT). Hence, you can selectively include or exclude channels/trials and time segments from further computations.
A data scoring facility enables you categorize different segments of your data, e.g. adding score REM sleep to EEG traces displaying rapid eye movement activity. Data scores can be loaded and saved for the specific data set.
Fourier Transformation and Band Power distribution
The data editor lets you easily investigate the power spectral density of selected signal segments during the review of your data set.
Simply select the time segment of interest by using the Epoching tool and click on Analyze.
The figure to the right displays the power spectral distribution for the selected EEG time segment. In this example, prominent rhythmic activity with high amplitude in the lower beta band is apparent.
The lower part of the figure displays the power contribution of the individual frequency bands. Hence alpha rhythmic activity, mu-rhythmic activity or theta and delta activity can be easily identified.
A measuring tool, which lets users measure e.g. peak amplitude and peak frequency, completes the toolkit.
Publication ready plots with the latest algorithms
The method of Independent Component Analysis (ICA) separates statistically independent source signals that have been mixed linearly into distinct output signals. In contrast to Principal Component Analysis, ICA finds temporally independent components even if they have similar scalp distributions.
One application of ICA is EOG artefact reduction and correction. The ICA output yields the time course of EOG source signal extracted from scalp EEG. The EOG signal can then be eliminated in a further processing step, revealing the "cleaned" EEG for further processing and analyses.
Powerful Batch Processing and user-defined algorithms
Data set processing steps and necessary parameter settings are typically performed with a mouse-click in the GUI.
However, once the processing steps are fixed, then group study data can be processed automatically in g.BSanalyze Batch mode.
Furthermore, all computation steps are well documented in a journal file, making it easy to follow the processing chain step-by-step.
[...] %Filter Filter.Realization='fft'; Filter.Type='BP'; Filter.Order=0; Filter.f_high=13; Filter.f_low=7; TrialExclude=; ChannelExclude=; P_C=gBSfilter(P_C,Filter,ChannelExclude,TrialExclude); [...]
- Data visualization in channel x trial and trial x channel mode
- Data ruler
- Undo function
- Journal file
- High speed data scrolling
- Data attributes and markers
- Epoching tool to score data
- Assign comments
- Jump to attributes
- Replay data
- Zoom in/out
File /IO and Printing
- Subject/Session information
- Import filters for many file formats
- Export ASCII data
- Load/Save scoring data
- Cut samples/trials/channels
- Sort data
- Merge data-sets
- Arithmetic operations (NEW abs)
- Data triggering
- Untrigger data
- Source derivation
- DC correction
- Smoothing & rectifying
- Data detrending
- Remove drift
- Data filtering (HP, LP, BP, Notch)
- Spatial filtering (CSP, ICA, PCA,...)
- NEW Moving average
- NEW Baseline correction
- Reaction time
- Single trial analysis
- NEW Trigger finder
- Overflows, zerolines, events
gtec's interactive environment for spatio-temporal EEG data processing
Specialized functions for EEG artifact detection and data quality computations ensure the highest-quality results. Adaptive filtering or bandpower analyses extract a variety of EEG signal parameters. The parameter extraction section also includes the extraction of Hjorth and Barlow parameters.
Signal similarities in the time domain are determined by cross-correlation methods. In the frequency domain, coherence analyses yield linear couplings as functions of frequency. Signal averaging methods allow determination of phase-locked Evoked Potentials. A section with sophisticated functions for Event-Related Desynchronization and Event-Related Synchronization analyses allows investigation of stimulus- but non-phase locked EEG phenomena.
The graphical user interface for g.EEGtoolbox enables you to explore EEG data sets and extract relevant information that cannot be seen from raw data only. Results of multi-channel computations are visualized in the result2D tool, which also allows topographical arrangement of the results. You can load and save your preferred processing steps with scripts and automatically process your data in batch mode.
- Quality Control (min, max, average, entropy,...)
- NEW Evoked Potentials (ASSR, MMN, BAEP, P300, N400,...) including statistical analysis
- Power spectrum estimation and comparison with significance analysis
- Wavelet analysis
- Bandpower estimation, cross correlation, running fractal dimension, temporal and spatial complexity, Hjorth parameter estimation, Barlow parameter estimation, AAR parameters for motor imagery based BCIs
- Variance for Common Spatial Patterns
- Exponential window for slow cortical potentials
- Cross correlation based template matching
- NEW Minimum energy for SSVEP/SSSEP based BCIs
- Source derivations (Laplacian, bipolar, CAR, Hjorth,...)
- Montage creator for topographic information
- Event-related coherence / Coherence analysis
- Event-related desynchronization (ERD) / event-related synchronization (ERS) analysis
- Time-Frequency ERD/ERS analysis with bootstraping, complex demodulation, Hilbert transformation and wavelets
- Common Spatial Patterns (CSP)
- Mean frequency and center frequency
- NEW Phase locking value (PLV)
- NEW Compressed spectral array
- NEW P300 BCI accuracy
The average function allows you to calculate EPs from 2 classes (target versus non-targets) and shows the results for all EEG channels in g.Result2D. You can also perform a statistical analysis to find significant differences between the 2 classes with different background colors. This makes the interpretation of the EPs easier.
g.ECGtoolbox is a software package for ECG data processing and analysis. The investigation of patterns and signal features of ECGs allows users to noninvasively observe brain and heart functions and dysfunctions.
The graphical user interface for g.ECGtoolbox enables you to investigate all important time and frequency domain features of your electrocardiogram (ECG) data, such as RR intervals or HRV maps. You can load and save your preferred processing steps as a script program and automatically process your data in batch mode.
QRS complex detection
|Starting from noisy raw ECG data, QRS complexes are automatically detected and indicated in the data editor with attribute QRS. An intelligent algorithm assigns the attribute QRSBAD to R-peaks that are not detected adequately. These markers can now be corrected visually or (if necessary) excluded from further analyses. Time courses of tachogram and RR-intervals can also be displayed for visual inspection.||
Time Domain ECG Features
Frequency Domain ECG Features
ECG data analysis in the frequency domain allows you to investigate heart rate variability oscillations at different frequencies. These oscillations are caused by different physiological systems. The parasympathetic and sympathetic systems modulate the heart rate variability. High frequency oscillations (about 0.2-0.35 Hz) are vagally mediated and low frequency oscillations (around 0.1 Hz) are due to both parasympathetic and sympathetic systems. The respiratory sinus arrhythmia (RSA) is vagally mediated and has a frequency synchronous with the respiratory cycle (between 0.2 and 0.35 Hz). Very low frequency components are associated with slow regulation mechanisms such as humoral and thermoregulation factors.
Time-Frequency ECG Analysis
Example: HRV maps
To simplify the data analysis and interpretation of the ECG data, HRV maps can estimate the PSD (power spectral density) for a certain segment. Then the segment is shifted by a specific step size, and the PSD is calculated again. This yields a comprehensive time-frequency analysis plot over the recording time. High power values are indicated in red, and low power values are shown in blue.
In addition to the time frequency plot, the time courses of the spectral components ULF, LF, HF, VHF are also displayed in the protocol.
- Robust QRS complex detection
- Overlayed QRS detection result with raw ECG
- Edit and manipulate QRS detection result
- Heart-rate variability measures in time domain (RMSSD, pNN50, SDSD,...)
- Geometric measures: HRVindex,...
- Absolute measures of spectral power distribution (ULF, VLF, LF, HF)
- Relative measures of spectral power distribution (LFnorm, HFnorm, LF/HF)
- HRV time-frequency maps
- NEW Event-related heart-rate and heart-rate variability
gtec's software package for linear and non-linear data set classification
Classification enables categorization of patterns and signal features of biosignals into different classes.
In the learning phase of supervised learning, classifiers are trained with signal patterns and signal features with known class information. After the training phase, the classifier is ready to categorize new signal patterns into the specific classes.
One application of classification is to discriminate EEG patterns in brain-computer-interface experiments (e.g. into LEFT and RIGHT hand motor imageries). Optimized feature selection can be performed via Distinction Sensitive Learning Vector Quantization (DSLVQ). This method yields the most relevant features for optimal data set classification.
The graphical user interface for g.CLASSIFYtoolbox helps you classify signal patterns of whole trials, within specified time segments or at certain time points. Signal features are combined in a so-called "feature matrix". This matrix is the basis for training all linear and non-linear classification methods, including a multi-class linear discriminant analysis, radial basis function neural network or multi-layer perceptron neural network. You can load and save your preferred processing steps as a script program and automatically process your data in batch mode.
The classification toolbox contains also statistical analysis functions to solve the zero class problem of BCI systems. This means the classifier can detect whether the person wants to select a certain control, and does not make a decision otherwise.
- Generate feature matrix
- Multi-class linear discriminant analysis
- Minimum distance classifier
- Backpropagation neural network
- Receiver operating curve
- Radial basis function neural network
- DSLVQ (distinction sensitive learning vector quantization)
- DSLVQ for feature weighting
- KMEANS clustering
- NEW Support Vector Machine (SVM)
- NEW Change rate and majority weight
- NEW Zero-class integrated
- NEW Classification output mapping
The classification output mapping functions shows the classification result of a motor imagery BCI experiment with right and left hand movement imagination in g.Result2D. The solid lines are the averages over all movement imaginations for a certain class.
gtec's interactive environment for high resolution EEG data processing
The investigation of patterns and signal features of EEG data, combined with anatomical information based on MRI or CT, data allows noninvasive observation of brain functions and dysfunctions.
With the graphical user interface of g.HReeg, you can perform spatiotemporal analyses of multi-channel electroencephalogram data. 3D spatial highpass filtering allows you to visualize local EEG phenomena e.g. in the central areas. In addition, you can load and save your preferred processing steps as a script program and automatically process your data in batch mode.
Combine EEG and anatomical data
Starting from segmented MR images, boundary models for scalp and brain are automatically generated. In order to see through the scalp, the transparency and the smoothness of the brain and scalp boundary models can be chosen. EEG maps are displayed on the surface of the scalp model or calculated as high resolution maps on the brain.
The high resolution EEG maps are calculated based on a 3D spline surface Laplacian method. The head is represented by a spline model and this geometric information is used in calculating the 2nd spatial derivatives. Smoothing of the results is done via truncated singular value decomposition.
Display Data Maps
|Display sequences of spatially highpass filtered data with the 3D spline surface Laplacian method on the surface of the head, at idealized planar electrode positions or on the surface of the brain.||
|Freely rotate the head model and change the view angle for optimal presentation, and change color maps to optimally display spatiotemporal information in the maps.||
gtec's software package for Cerebral Function analysis
With this toolbox, you can calculate the amplitude integrated EEG (aEEG) signal from any already recorded EEG signal. The aEEG signal is also known as cerebral function monitor signal (CFM) and is used in the neonatal intensive care unit. The aEEG signal gives a time compressed activity report of the ongoing EEG signal and has gained widespread popularity as an alternative to the conventional EEG monitoring in neonates. aEEG monitoring is very useful for the assessment of ischemic encephalopathy, seizures and intraventricular haemorrhage.
The aEEG signal
For any EEG channel and specific intervals, the toolbox can calculate the aEEG signal. Therefore, the EEG data is first filtered by an asymmetric bandpass filter to compensate the 1/f damping of the EEG signal. This filter enhances signal components with higher frequencies. Normally, the lower cut-off frequency is 2 Hz and the upper cut-off frequency is 15 Hz. Then each sample value is rectified and averaged over a time interval - this is the aEEG or CFM signal. The aEEG is plotted between 1µV and 10 µV on a linear and between 10 µV and 100 µV on a logarithmic scale. Therefore, lower amplitudes are easier to interpret. Traditionally, the aEEG is plotted as 6 cm per hour. Thus, a whole night can be visualized on only 3 A4 pages, which helps facilitate later investigation.
The toolbox can automatically classify the aEEG signal into specific patterns. Therefore, the aEEG signal is divided into segments of e.g. 10 minutes. The maximum of each interval is defined as the 95 % percentile and the minimum as the 5 % percentile and both are indicated by the red horizontal line in the figure above. Patterns can then be assigned to different segments, as defined in the article below published by Guger, Klebermass and Olischar:
A new and improved classification algorithm was implemented to assign these patterns automatically. Initially, 10 training subjects were scored by experts and then the new algorithm was trained to classify as accurate as possible the aEEG traces. After reaching an accuracy of >80% the algorithm was applied on CFM signals from 10 testing subjects (S11-S20). The testing subjects were also scored by the experts.
The colors indicate the patterns assigned to the aEEG signal.
The table shows the overlap between the automatic classification and the experts' classification. 77 % overlap was achieved for a segment length of 10 min for all subjects. 80.9 % overlap was attained with 5 min segmentation.
Notably, even unseen data show high overlap with the classification done by experts (>80 %). The new method can easily be used in practice and improves and simplifies the aEEG annotation. This supports also the fast interpretation of the CFM signals. The expert immediately identifies the interesting parts of the CFM trace and can more easily investigate these parts in detail.
C. Guger K. Klebermass, M. Olischar, G. Edlinger, Automatic classification of amplitude integrated EEG patterns, Biomedical Engineering, submitted, 2008.
- Calculation of the aEEG signal of any recorded EEG channel
- Automatic classification of the aEEG signal into distinct patterns
- Raw EEG activity can be investigated together with the aEEG signal
- Toolbox allows the off-line analysis of the calculated aEEG data
- Toolbox supports group studies
Acquisition and Analysis of fetal ECG
Task #1: Record and analyze fetal ECG
Record 8 channels of ECG data from the mother to find the fetal ECG. Independent Component Analysis (ICA) should be used to find the sources of the fetal signal.
Perform the following steps:
- Place 8 ECG disposable electrodes on the mother's abdomen, where you think the baby's hear is. Arrange the 8 electrodes like shown in the MontageCreator below:
The electrode distance should be about 4-5 cm.
- Connect g.USBamp to the notebook and start g.Recorder for data acquisition (or use g.HIsys). Connect the 8 electrodes shown in the MontageCreator to inputs 1-8 of the amplifier.
- Place the ground electrode on the mother's foot and connect the cable to the G-Ground electrode socket of the amplifier.
- Place the reference electrode into the right lower corner of the 3 x 3 grid shown on the MontageCreator and connect it to the R-Reference input of the amplifier.
- Start g.Recorder and select 256 Hz. Check the data quality of the ECG channels and record about 90 seconds of data. Use the filename FetalECG.hdf5.
- Load the data-set into g.BSanalyze to investigate the 8 channels:
In the raw data, neither the mother's nor the baby's ECG are visible without signal processing techniques because noise overrides the very small signals.
- Open the Geometry window from the Header menu and Browse for the 8chmontage.mat file.
This loads the electrodes x- and y- coordinates into g.BSanalyze. This information is important for the ICA calculation in the next step.
- Open the Independent Component Analysis (ICA) window from the Analyze menu and select higher order statistics (HOS) under ICA algorithm.
- Check Save results and select the filename icafilter.mat to store the result of the analysis.
- Open the Spatial Filter window from the Pre-Processing menu and load the icafilter.mat file that was calculated in the previous step. The filter has a size of 8 x 8 because it was calculated with 8 ECG channels.
- Select the Create temporal pattern function and enter (under use filter(s) with number(s) [1:8]) to use all 8 ICA filters.
- Select Replace all channels to remove the raw ECG data. Press Start to perform the Spatial Filtering.
- Finally, the DataEditor shows the ICA components of the 8 ECG channels and contains the ECG of the mother in channel 7 and the ECG of the baby in Channel 3. The other channels contain mainly noise (especially channel 4 contains the power line interference noise). Note the small amplitudes in the µV ranges of the data.
Task #2: Find QRS complexes and calculate HRV
Assign QRS markers to the ECGs of the mother and of the baby. Calculate and compare the heart rate variability parameters in time domain.
- Click on the Select button in the Data Editor to define new markers.
- Create a marker with the name FETALQRS and another marker with the name QRSmother. Close the window by clicking OK.
- Now search for the QRS peaks of the mother and assign the QRSmother marker to each peak. The markers are indicated in blue color in the figure below:
- Similarly, assign the FETALQRS markers to the ECG channel 3. The markers are indicated in red. NOTE: You can also load the file ECGwithMarker.mat to avoid the manual editing.
- To calculate the heart rate and heart rate variability parameters in time domain, open the HRV Time Domain function from the ECG menu.
- Select the QRSmother marker and set the Start interval to 5000 ms and End to 60000 ms. Also, set the Tachogram unit and Histogram unit to bpm. Check the Resample tachogram box, and then press Start. This calculates the Tachogram, the Histogram and the most important HRV parameters in the time domain of the mother.
- g.Result2d opens with the results:
The heart rate varies between 77 and 99 bpm. The histogram is quite narrow and the highest peak is 12. This means that 12 beats of the data segment had the same heart rate.
- Then perform the steps with the FETALQRS marker to obtain the following results:
The heart rate varies now between 120 and 174 bpm. The histogram is much broader and only 10 beats had the same heart-rate.The table below shows the most important parameters for the mother. The MeanHR is 90.01 bpm, the RMSSD is 23.94 ms and the pNN50 is 3.7%.For the baby, the MeanHR is 150.48, the RMSSD is 17.38 and the pNN50 is 0.75 %. This means that the baby has a higher heart rate and lower heart rate variability than the mother, which is common.
The spike toolbox contains specialized functions for the analysis of position, spike activity and multi-unit activity.
Multi-unit activity of neurons
The peri-stimulus time histogram allows you to visualize event-related activity of multi-units to identify the reactivity of certain areas e.g. of the cerebellum. The algorithm calculates the baseline and then only plots points which are higher than a certain threshold. The threshold is calculated from the baseline activity. This allows you to identify on-set and off-set times of multi-unit activity.
Position related physiological parameter analysis
The toolbox allows to load position information acquired with a video tracking system to calculate movement trajectories, speed and dwell times. This position information can be analyzed together with physiological parameters such as heart-rate, bandpower or spiking frequency to create activity maps. These maps identify e.g. places with low heart-rate and high heart-rate variability and could be considered as e.g. the home-base of a rat. If neuronal spiking frequency is used, firing fields can be calculated that identify place fields of cells in the hippocampus of rats. Furthermore, statistical parameters of the map can be calculated, such as maximum/minimum/mean frequency, spatial coherence, spatial selectivity, mean non zero rate and Skaggs index.
- analyze position data
- map physiological parameters onto position data
- analyze place cells and place fields
Mapping Feature Channels to Position – Spike Activity
Usually, spike data is not recorded as a raw waveform, because the data between spikes does not need to be stored. This means that a lot of bandwidth and memory would be wasted. Therefore, a threshold is set to each recording channel of an experiment. If the signal crosses this threshold from below, about 10 ms of data are stored around the time point the threshold was crossed. The time of the event is also stored. The cut windows are then used to assign the spikes to neurons; that is, to discriminate spike shapes with the assumption that they were produced by different units distributed around the recording electrode. Some events may not be assigned to units. After this process, the spike data is represented by two very compact lists:
- Spike time: The time of the event relative to the start of the recording.
- Neuron ID: The number of the neuron assigned to the event as a result of the spike sorting process.
|Spike time||Neuron ID|
|0.19||1||At time 0.19 neuron 1 fired.|
|1.04||4||At time 1.04 neuron 4 fired.|
|1.13||2||At time 1.13 neuron 2 fired.|
|1.19||3||At time 1.19 neuron 3 fired.|
|2.35||2||At time 2.35 neuron 2 fired.|
|2.38||2||At time 2.38 neuron 2 fired.|
|3.17||-1||At time 3.17 there was an unknown event.|
|3.89||4||At time 3.89 neuron 4 fired.|
If the spikes are recorded along with position of the animal (video tracking system) then firing maps of place cells in the hippocampus can be calculated. The figure below shows x- and y-positions of a rat running in a box measuring 0.8 x 0.8 m. Spike activity is recorded from 4 hippocampal neurons.
The Spike Toolbox allows calculating the movement trajectories, the visits of each class (pixel) of the arena, the probability that the rat was at a certain pixel and the dwell time.
Then the spiking activity can be mapped onto positions, as shown in the figure below. The first two rows show very nice place cells with distinct borders. This means that these 6 neurons only fire if the rat is at a certain position; otherwise, the neuron is not active. The last row shows neuronal activity that is not related to position.
Mapping Position Data - example of a swinging pendulum
The g.BSanalyze Spike Toolbox lets users map a feature channel – in this example, the speed of a pendulum – in the same figure as the position of the swinging pendulum using gBSpositionmaps. The position and speed data of a moving pendulum were tracked by a video tracking system. The first two channels of the tracked data present the x- and y-coordinates of a pendulum while it moved over time. The third channel is the speed of the moving pendulum over time. The Data Editor shows the first 20 seconds of the recording.
The x data is in a range between 0 and 1.1 meters. The y data is in a range between 0 and 0.2 meters. The third channel represents the speed of the pendulum as it swung.
After calculation, g.Results2D opens with the position trajectory and the rasterized data. In the picture below, the first plot shows the position trajectory. The second plot shows the visits per class (the number of samples the position was rasterized to this class). The third plot shows the class probability (the visits per class divided by the total number of visits from all classes). The fourth plot shows the dwell time (the visits per class divided by the sampling frequency).
Mapping Features Channels to Position – Pendulum
In this example, we will analyze another feature of the pendulum. After opening the tracked data file and calculating the activity map, the feature channel is plotted. The color represents the amount of activity, i.e. the average speed in each class. As expected, the average speed is a bit higher when the pendulum is at its lowest point, since the pendulum changes from falling to climbing. The maximum speed is about 0.7 m/s.
Because the recording is relatively long and continues until the pendulum almost stopped moving, the mean value of the speed in each class is lowered by samples of the feature from a later period of the recording. It is possible to exclude this data and the calculation interval can be set. After recalculating, the maximum speed is now 0.8 m/s instead of 0.7 m/s.