Most BCI platforms use a single EEG paradigm, which may not be suitable for all users and can also produce erroneous identifications. In recent years, BCI systems have been enhanced by using multiple paradigms to improve the accuracy and speed of controlling external applications.

Researchers found through extensive research that previous studies have confirmed the possibility of simultaneously detecting P300 components and SSVEP activity. Most hardware for P300 and SSVEP stimulation is based on computer screens, which are limited by screen refresh rates, leading to lower portability and causing visual fatigue in participants.

In this study, the researchers designed a fully portable hybrid system to evoke P300 and SSVEP using a high-precision dedicated hardware platform. In this hybrid hardware, SSVEP is used as the primary response, while P300 serves as a corrective mechanism in classification. In this study, radial green LED stimuli on four independent chips were used to evoke SSVEP, which was controlled separately by a microcontroller platform to produce precise flashing frequencies.

Figure 1.1 Hybrid BCI platform based on SSVEP and P300

The stimulation platform is controlled by independent control systems, as shown in Figure 1.1. The four radial green stimuli used to induce SSVEP are controlled by four Teensy 32-bit microcontroller modules, while the four red stimuli for P300 are controlled by separate Teensy modules. Teensy also sends the timestamps of each flash event to the recording software.

Hardware Description

For multi-SSVEP stimulation, four independent Teensy microcontroller platforms are used to produce flashing frequencies of 7, 8, 9, and 10 Hz under green radial stimuli. High-power red LEDs are placed within each radial ring to evoke P300 events and are tagged as events along with SSVEP EEG data. The timing of the flashing is controlled randomly by a single Teensy module. Flashing events are transmitted as serial data from the microcontroller to the EEG recording software.

Researchers found in a study that the duty cycle of each flashing frequency of the precisely generated SSVEP is 85%, as this duty cycle can provide the highest performance. Therefore, the researchers used high-power MOSFETs (A09T) driven by a shut-off regulator MP1584 to drive LED stimuli, providing a constant current source of 3A, ensuring optimal brightness throughout the experiment. The design of the regulator is shown in Figure 1.2.

Figure 1.2 Regulator design for visual stimulation

The entire hardware is powered by a 12 V 10A battery power supply to avoid any power interference. The complete schematic of the SSVEP radial stimulation is shown in Figure 1.3.

Figure 1.3 SSVEP radial stimulation design

In this study, four modules with the same design were used, and different flashing frequencies were programmed for classification. Firmware was developed to generate the flashes, and a digital oscilloscope was used to verify the accuracy of the flashes. The waveforms for the four frequencies are shown in Figures 1.4 and 1.5. The stimulation positions and layout are shown in Figure 1.6.

Figure 1.4 Stimulation frequencies of 7 Hz and 8 Hz with a duty cycle of 85%

Figure 1.5 Stimulation frequencies of 9 Hz and 10 Hz with a duty cycle of 85%

Figure 1.6 Placement of mixed stimulation LEDs

To evoke the P300 component, four random flashes were generated using red LEDs, and the flash event timestamps were sent separately to the data recording software. Serial communication (Rx and Tx) was used to transmit the event timestamps from the microcontroller to the computer. The TTL levels from the microcontroller were converted to RS232 using MAX3232.

As mentioned earlier, the red LED driver circuit was designed for high current. The complete schematic for the P300 flasher and event markers is shown in Figure 1.7.

Random flashing timing of the four red LEDs sends timestamp values to the EMOTIV testbed software. The LEDs located within the radial ring are marked as 111, 112, 113, and 114 for 7, 8, 9, and 10 Hz, respectively, and these values are stored as separate channels in the EEG data during recording. Figure 1.8 shows the Testbench software with EEG data and marked events. The random flash timing is set between 200 to 800 milliseconds. The baud rate values for serial communication on both the transmitter and receiver sides are set to 115,200.

Production Instructions

Hardware Assembly

The system includes two types of stimulus designs, one for SSVEP evocation and the other for P300 evocation. For SSVEP, there are four separate modules that precisely generate four different frequencies of 7, 8, 9, and 10 Hz. Each Teensy module is programmed with the developed firmware to achieve the desired flashing frequency. The output of the MP1584 regulator is set to approximately 10.6 V DC and connected to the radial stimulator as shown in Figure 1.3. Similarly, the other three Teensy modules were programmed and wired.

Figure 1.7 P300 stimulation and event markers

For P300, as shown in Figure 1.7, the red LEDs are connected to the four output pins of the Teensy module, and the LEDs are placed at the center of each COB ring. The output of the MP1584 needs to be set to 2.8 V DC for optimal brightness of the red LEDs. For serial communication, the pin 1 of the Teensy module (Tx) needs to be connected to pin 13 of MAX3232, which is the serial data receiver (Rx). The Testbench software and event markers can be viewed as shown below.

Figure 1.8 Testbench software and event markers

Programming

The Teensy modules can be programmed directly via USB using the open-source Arduino IDE. The IDE installer can be downloaded from the internet (www.arduino.cc), which is needed to load the firmware. Each small module needs to be programmed separately with the firmware required for the selected stimulus position. Starting from the top COB LED stimulator, the prototypes are programmed in a counter-clockwise direction with 7, 8, 9, and 10 Hz.

Prototype

The prototype is shown in Figure 1.9. To avoid light reflection from the A3 size (21 cm x 29.7 cm), the substrate is made of acrylic with a black matte surface. The P300 LEDs are fixed at the center of the ring using double-sided tape and connected to the hardware on the back of the board. The radial COB ring is soldered with fine wires for connection to the MOSFET driver. The ring is fixed in the position as shown in the following figure using double-sided tape.

Figure 1.9 Prototype

The control hardware is built on a universal PCB that has 5 28-pin IC sockets, with components connected using single-strand wire according to the schematics in Figures 1.3 and 1.7.

The EEG recording process starts with a 7 Hz stimulus for 3 seconds, followed by a five-second break when the participant shifts their gaze away from the flashing stimulus. This is followed by stimuli at frequencies of 8, 9, and 10 Hz, with the same break time, to complete a full session. Each participant recorded five sessions containing EEG data for SSVEP and P300 events. The timing of the stimuli for each frequency and rest period is shown in the figure below.

Figure 1.10 Stimulation timing diagram

The 3 seconds on the timing diagram represent the time participants focus on each stimulus, and the five seconds represent the rest time between the transition from one stimulus to the next. The EEG data recorded in EDF format is converted to MATLAB format using EEGLAB. A program is then written in MATLAB to process the data from SSVEP and P300 event detection to evaluate the mixed stimulation.

The following figure shows a photo of using this hybrid BCI for LEGO control.

Figure 1.11 Controlling LEGO with hybrid BCI

Data Analysis

For SSVEP analysis, data is extracted from channel O2 during the required three-second time period at each frequency and stored for further processing. For SSVEP data classification, the main frequency and its first and second harmonics are analyzed. The stored EEG data is filtered with a band-pass filter with a center frequency of the stimulation frequency and a bandwidth of 2 Hz. This is performed for all main frequencies and their harmonics. The variance of each filtered data is calculated (which represents the energy of the signal), including the main frequency and adding the two harmonics, and stored for classification analysis.

For P300 analysis, data is extracted from channel F4 along with the four event markers. The event markers are set to 111 for 7 Hz, 222 for 8 Hz, 333 for 9 Hz, and 444 for 10 Hz. For the same time range as in the SSVEP analysis (the same time range before the participant focused on the stimulus), the algorithm checks the event markers to find peaks within 600 milliseconds of the event markers. This is performed for all four markers at all four frequencies. The peaks for all frequencies in the required time window are stored for classification accuracy analysis.

Results

The researchers explored the possibility of combining two EEG paradigms to develop hybrid BCI visual stimulation hardware that can reduce the focus time of visual stimuli and allow good single-trial classification output. The developed independent mixed stimuli successfully generated frequencies of 7, 8, 9, and 10 Hz, with small gaps between them. P300 events were also generated simultaneously with the four event markers and successfully detected in the recorded EEG using MATLAB. From the data analysis of five participants, SSVEP and BCI produced better classification rates within the same time window for each frequency. P300 also showed good performance.

Additionally, the researchers successfully achieved command control based on SSVEP to control the movement of a LEGO robot using mixed visual stimuli (which can be used for severely paralyzed individuals to control wheelchairs in the real world). Four low-frequency flashing visual stimuli were presented for the user to focus on one at a time to execute movement commands for the LEGO robot. For each visual stimulus, the command was sent to the robot with a minimum time of 3 – 4 seconds. Besides a slight delay in data processing and classification, the robot’s movement is almost real-time. The slight delay for each action is about 3-4 seconds, which is due to the data processing time required before executing the action.

The researchers stated that this hardware platform could serve as an independent visual stimulation device or be used with event markers for P300 event-related experiments for various neurological investigations or psychological research. The device can be easily customized for single or multiple stimuli.

Reference Links:

DIY hybrid SSVEP-P300 LED stimuli for BCI platform using EMOTIV EEG headset

More readings

Summary of public datasets on motor imagery, EEG emotions, etc.

Can money buy a better brain?

Researchers call for attention to the issue of neural data privacy

Detailed explanation of EEG artifact types and summary of filtering tools (1)

EEGLAB tutorial series (10) – Drawing ERP images

Breaking! For the first time, brain-computer interfaces allow patients to output complete sentences

Data superposition average {2} draws 2D and 3D graphs

Detailed explanation of EEG artifacts and summary of filtering tools (2)

Submission Channel

How to make your work known and beneficial to more people?

The brain-computer interface community is a platform channel that connects academia,

industry, and enthusiasts.

The community encourages universities, businesses, or individuals to share quality content on our platform.

Submission Requirements

The submission is an original work by individuals; if published on other platforms, please clearly indicate.

Once accepted, a manuscript fee will be provided!

Submission Channel

Scan the WeChat QR code, note: submission + name + unit

WeChat group, please scan the above WeChat

(Note: name + unit + major/field/industry)

QQ group: 913607986

Your one view can make me happy all day!