In recent years, Brain-Computer Interface (BCI) technology has developed rapidly and has become a research hotspot in the field of neuroscience and engineering. Steady-State Visual Evoked Potentials (SSVEPs) have been widely used in BCI systems as a reliable EEG signal that elicits a stable EEG response through visual stimulation. However, the traditional SSVEP-BCI system is limited by low frequency resolution and has a low information transmission rate.

In this paper, we propose an SSVEP-BCI system based on High-Density Frequency Division Multiplexing (HDFDM), which greatly improves the information transmission rate by encoding and transmitting information at the same time at hundreds of frequencies, and demonstrates the application potential of the system in image transmission and simple classification tasks.

Materials and methods

Experimental equipment

A custom LED light source system and EEG equipment were used in the experiment. LED light sources are used to generate light signals of different frequencies, and EEG devices are used to record the subject's visual cortex electrical signals. The configuration is as follows: LED light source: including multiple groups of LED lights that can be controlled independently, and each group of LED lights can adjust different modulation frequency and amplitude. These light sources are computer-controlled, and the frequency and amplitude of each LED are precisely adjusted through pre-programmed programs. EEG device: 3-pole electrode configuration, with the active electrode in the Oz position (the center of the back of the head), the reference electrode above the left ear, and the ground electrode above the right ear. The EEG signal is amplified by an amplifier and digitized by an analog-to-digital converter (ADC).

High-density frequency-division multiplexing SSVEP

The encoding and decoding process of high-density frequency-division multiplexed SSVEP is demonstrated. By using hundreds of frequencies, multiple pieces of information are transmitted simultaneously and the SSVEP signal is recorded via EEG devices. Figure 1 illustrates the experimental setup and signal processing flow, including encoding of input data, LED light signal projection, EEG recording, and spectrum analysis.

Figure 1 | BCI settings. The input data (example image is a handwritten number "0" and a set of control parameters) is encoded by frequency division multiplexing. The frequency-coded signal modulates the intensity of the LED light projected onto a white screen, as observed by the participant. A 3-pole EEG device detects steady-state visual evoked potentials in which an active electrode is placed at Oz (medial occipital electrode site) to capture electrical signals from the primary visual cortex, with a reference electrode above the left ear (M1 position) and a grounding electrode above the right ear (M2 position). The resulting normalized power spectral density (NPSD) is used for image transmission or computational tasks.

Frequency-division multiplexing coding

In an experiment, the input data can be an image (e.g., a handwritten number) or a set of parameters. With high-density frequency-division multiplexing encoding, each image pixel or parameter is assigned a specific optical modulation frequency (fm) and its corresponding amplitude (Am). These combined signals of frequency and amplitude are projected onto a white screen via LEDs. Subjects watched these light signals and recorded the SSVEP signal in the visual cortex by an EEG device. The first is data preprocessing, which preprocesses input data, such as images or parameters, into a format suitable for frequency division multiplexing coding. For images, the grayscale value of each pixel is normalized and mapped to the corresponding light frequency and amplitude. Frequency assignment is then carried out, assigning a unique light modulation frequency (fm) and its amplitude (Am) to each data unit (e.g., pixel) based on the preprocessed data. These frequencies and amplitudes should be selected to avoid inter-frequency interference and ensure the accuracy of signal transmission. Finally, signal generation takes place: according to the assigned frequency and amplitude, the optical signal is generated by the LED light source system. The signal generation process is controlled by a computer, ensuring that each LED light emits light at a predetermined frequency and amplitude.

Image Transfer

To test the system's ability to transmit information, a 14×14-pixel handwritten digital image was selected. First, image preprocessing is carried out: the image pixel matrix is linearly expanded into a one-dimensional vector, and the gray value of each pixel is used to modulate the amplitude of the corresponding optical frequency. Then there is the light signal generation: the modulated light signal is projected onto the screen via an LED light source system. After that, the signal is recorded: the subject looks at the screen, and the EEG device records the SSVEP signal. Then there is signal processing: the recorded SSVEP signal is analyzed to obtain a Normalized Power Spectral Density (NPSD). Finally, the image is reconstructed: the amplitude of the frequency component in the NPSD is used to reconstruct the original image.

Physical neural networks

In order to demonstrate the computing power of SSVEP-BCI, a Physical Neural Network (PNN) based on SSVEP was constructed. The network leverages the nonlinear response of the vision system to blend input data with control parameters to achieve a simple classification task. The specific steps are as follows:

1. Input data: Select an 8×8 pixel handwritten digital image for classification.

2. Optical signal generation: The image pixel value is converted into an optical signal and projected onto the screen through the LED light source system.

3. Signal recording: The subject looks at the screen, and the EEG device records the SSVEP signal.

4. Signal processing: Perform spectrum analysis on the SSVEP signal to extract the amplitude of the frequency component.

5. Classification tasks: Optimize classification performance by adjusting control parameters.

Experimental procedure

First of all, the experimental setup consists of an LED light source system, an EEG device and a data processing system. The input data is encoded by a specific algorithm and converted into an optically modulated signal. These light-modulated signals are then projected onto the screen via an LED system to form a visual stimulus that the subject needs to see. While subjects watched these light signals, the EEG device recorded their steady-state visual evoked potentials (SSVEPs). Finally, the recorded SSVEP signal was analyzed in the spectrum, and the amplitude of each frequency component was extracted for subsequent signal processing and decoding. This series of steps ensures the accurate acquisition and processing of experimental data, and provides a basis for verifying the effectiveness of high-density frequency-division multiplexing SSVEP technology.

Experimental results

The experimental results show that multiple information can be transmitted at the same time through the high-density frequency division multiplexing technology, which significantly improves the information transmission rate. Figure 2 shows the SSVEP spectrum and reconstructed images at different bandwidths. The results show that with the increase of bandwidth, the noise of image reconstruction gradually decreases, and the image quality gradually improves.

Figure 2 | BCI image transmission. The results of the experiment showed an image of the handwritten number "7". The first row shows the normalized power spectral density (NPSD) of the steady-state visually evoked potentials generated by frequency division multiplexing, following equation (1), where f0 = 12 Hz, measurement time 196 sec, bandwidths of a1 Hz, b 2 Hz, c 4 Hz, d 8 Hz, e 12 Hz, f 16 Hz, respectively. g is the 12 Hz bandwidth and a shorter measurement time of 16.3 seconds, while h is the 12 Hz bandwidth and wearing an eyecup (only the α crest at 10 Hz is shown), and i is a 12 to 13 Hz magnification map. In the second row, J-Q shows the reconstructed grayscale image corresponding to the data above the image in the first row. Each plot also shows a structural similarity index measure (SSIM) to the real image, as shown in figure (r).

Image transfer experiments

In the experiment, the image transmission effect under different bandwidths was tested. The band widths from 1 Hz to 16 Hz correspond to acquisition times from 196 s to 16.3 seconds, respectively. The specific results are as follows: low bandwidth: under the bandwidth of 1Hz, the acquisition time is 196s, the image reconstruction effect is poor, and the noise is large. Medium bandwidth: At 8Hz bandwidth, the acquisition time is 24.5s, and the image reconstruction effect is improved, but there is still some noise. High bandwidth: At 16Hz bandwidth, the acquisition time is 16.3s, the image reconstruction quality is high, and the noise is significantly reduced. Figure 2 shows the SSVEP spectrum and reconstructed images at different bandwidths. The results show that with the increase of bandwidth, the noise of image reconstruction gradually decreases, and the image quality gradually improves. In addition, the experiment also shows that the acquisition time is shortened and the image reconstruction quality does not degrade significantly at a wider bandwidth.

Physical Neural Network Classification Experiments

In order to verify the computational power of SSVEP-BCI, the classification task of handwritten digits "0" and "1" was carried out. The first is to enter the data: choose an 8×8 pixel handwritten digital image. Light signal generation is then carried out: the pixel values of the image are converted into light signals and projected onto the screen via an LED light source system. After that, the signal is recorded: the subject looks at the screen, and the EEG device records the SSVEP signal. Then there is signal processing: the SSVEP signal is analyzed in the spectrum to extract the amplitude of the frequency component. Finally, the classification task is carried out: the classification performance is optimized by adjusting the control parameters. Figure 3 illustrates the experimental process and results. The results show that the SSVEP-BCI can efficiently perform simple classification tasks by selecting the frequency and control parameters appropriately. This shows that SSVEP-BCI has good scalability and application potential.

Figure 3 | BCI Physical Neural Network Image Classification. Experimental results of a single classification experiment with the handwritten numbers "0" and "1". a An example of input data, a number "0" with a gray scale of 8×8 pixels. b Measured EEG signal NPSD with three highlighted frequency intervals: the input image is frequency coded as 64 equidistant frequencies in the [15.0, 15.5] Hz range; The control parameters (determined by the genetic algorithm) were frequency coded as 64 frequencies in the range of [20.0, 20.5] Hz; and 128 intermodulation frequencies in the [35, 36] Hz range. c Decode intermodulation signals in more detail; The blue curve is the amplification of the measured signal in (b) and the red curve is the synthesized (numerical simulation) data. d Read probability distributions on the two categories "0" and "1", showing the correct classification of "0" (highest probability).

Further analysis

Through further analysis of the classification task, it is found that different frequency combinations and control parameters have a significant impact on the classification performance. Experiments show that appropriate frequency selection and parameter adjustment can significantly improve the classification accuracy. Further experiments also show the stability of PNN under different ambient light conditions, which shows that the system has good robustness in practical applications.

Figure 4 | Classification of single- and multi-layer physical neural networks. a Schematic architecture of a two-layer physical neural network (PNN). b Classification probabilities of single-layer PNNs applied to the iris dataset, which contains three categories. The correct classification is indicated by a gray bar. c Classification probability of a double-layer PNN applied to the same iris dataset. All three classifications are now correct, and the classification probability has increased significantly, from about 50% or less, to nearly 80%.

Figure 5 | Effect of attention on the frequency power of physical neural network classification and intermodulation (IM). a Physical Neural Network (PNN) classification probability (i.e., the fraction of power in each frequency band) as well as b Intermodulation (IM) frequency power, for a bilayer (brain) PNN with six participants, each participant acting only as a second layer (the first layer is fixed, participant 1). Participants were asked to "focus" (blue bar) attention on light flashing during light flashing (200 seconds) or to "interfere" (red bar) attention by performing mathematical operations (addition, subtraction, division). In all cases, participants look at the illuminated area on the screen. Each participant was measured twice, a few minutes apart, reversing the order of the "concentration" and "interference" conditions to rule out possible confounding effects from the chronological order of the condition execution. We found that PNN classification accuracy (t(5) = 6.29, p= 0.00006) and intermodulation frequency power (t(5) = 4.18, p= 0.002) were statistically significantly reduced compared to "centralized" conditions under the "interference" vs. "centralized" conditions. These results suggest that human attention can indeed directly alter the effectiveness of multi-layered brain connections and the computational efficiency of PNNs.

discuss

This paper demonstrates the possibility of encoding and transmitting information in SSVEP by high-density frequency-division multiplexing techniques. This technology not only improves the information transmission rate, but also demonstrates the potential of SSVEP-BCI in image transmission and simple classification tasks.

Future research can further optimize the frequency selection and signal processing algorithms to improve the performance and application range of the system. Specific optimization directions include:

1. Frequency selection: Improve the efficiency and accuracy of signal transmission by optimizing the frequency selection algorithm. Frequency allocation can be dynamically adjusted by machine learning algorithms to suit different application needs.

2. Signal processing: Develop more efficient signal processing algorithms to improve the extraction and analysis capabilities of SSVEP signals. For example, more advanced spectrum analysis techniques or deep learning models can be used to improve the recognition rate of signals.

3. Application Scenarios: Explore more real-world application scenarios, such as medical assistance, intelligent control, and entertainment. Virtual reality (VR) and augmented reality (AR) technologies can be combined to develop new human-computer interaction systems.

In addition, high-density frequency division multiplexing SSVEP technology has a wide range of application potential in several fields, including but not limited to:

1. Medical assistance: Provide more efficient means of communication and control for patients with movement disorders. For example, SSVEP-based augmentative communication devices can be developed to help patients interact with the outside world through visual control.

2. Intelligent control: Realize the precise control of complex equipment, such as wheelchairs, robotic arms, etc. It is possible to combine robotics technology to develop a smart home control system based on SSVEP.

3. Entertainment: Develop an immersive entertainment system based on SSVEP to improve user experience. It can combine the game and film and television industries to develop new entertainment products.

In this paper, the feasibility and effectiveness of high-density frequency-division multiplexing SSVEP are verified by experiments, and its application potential in information transmission and computing tasks is demonstrated. The results of this study provide new ideas and methods for the development of neural interface technology, which is expected to be widely used in the fields of assistive technology and cognitive enhancement.

It is only used for academic sharing, if there is any infringement, please leave a message and delete the infringement immediately!

This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com