Resources

1. Op-Amp Design Tool:

Use this spreadsheet to design a 2 stage Op-Amp from Specifications. This uses the method provided in Allen-Holberg (3e) and does not use the gm/Id method due to technology restrictions. Only Online Access is permitted (no reproduction/copy/download) – Please contact chatterjee.b@ufl.edu if you need to use it elsewhere.

2. Codes and Tools for Asymmetric Coil design for implants for optimum power transfer:

Relevant paper: Novel Systematic Design of Asymmetric Flexible Transceiver Coils with Optimal Wireless Power Transfer for Biomedical Implants; EMBC, 2024; Authors: Asif Iftekhar Omi and Baibhab Chatterjee.

The Rights to this version of the data and codes belong to WISE Lab, University of Florida (https://github.com/WISE-Lab-UF/Asym-TRX-Coils/tree/main).

3. Neural signal dataset with data augmentation for handwritten character recognition:

We used the neural signal dataset (ECoG) for handwritten character recognition from the paper [1]. In [1], the authors acquired ECoG signals from a patient who provided his neural signals while writing 31 characters and sentences in 10 separate sessions, with each character having 117 samples. In [1], the authors proposed a complex CuDNNGRU model for detecting the handwritten character from the neural data. We, on the other hand, wanted to perform inferencing on a portable platform, and hence, could not use any complex model. Because of this reason, we represented the raw neural signals data as images and used Alexnet and Resnet50 models on the raw dataset. However, the custom raw dataset exhibited overfitting on the Alexnet and Resnet50 models. Consequently, we used random noise injection and time-shifting-based data augmentation on the raw dataset, which makes the data 3 times larger than the custom raw dataset and prevented the overfitting challenges.

The dataaugmentation.py file is used for data augmentation.

Raw and augmented dataset link

The Rights to this version of the data and codes belong to WISE Lab, University of Florida

Reference:

[1] Willett et al., “High-performance brain-to-text communication via handwriting,” Nature, vol. 593, no. 7858, pp. 249–254, 2021.