Energy harvesting and application of deep learning in spatial modulation

Abstract

In this semester, We have worked on the following research problems: 1. Energy Harvesting ADCs: A common sample-and-hold (S/H) based ADC architecture tracks the analog signal for a fraction of the sampling period, and the hold period is irrelevant for these ADCs. Thus, we have proposed eSampling ADCs which utilizes the hold phase for harvesting energy from the analog signals without altering its analog to digital conversion procedures. To verify the feasibility of eSampling ADCs, a circuit-level design has been developed on standard CMOS 65nm technology. An 8-bit eSampling ADC which samples at 40 MHz has been designed on Cadence Virtuoso. Finally, an ultra low-voltage CMOS operational amplifier is designed to demonstrate the real-world usability of the harvested energy of the proposed ADCs. This semester’s work discusses the ( in Part I ) development of the mentioned op-amp. The proposed op-amp is a two-stage Millercompensated amplifier developed on the Cadence Virtuoso platform using standard CMOS 65nm technology. The op-amp operates with all transistors in the subthreshold region to achieve low voltage applications. The designed circuit achieves a DC gain of 66dB, a gain-bandwidth product of 383KHz, a phase margin of 58.4deg and total power consumption is sub-1-μW with a 0.56V voltage supply. Therefore, the design of low-voltage op-amp completes the main work of eSampling ADC. 2. Deep learning applications in Spatial Modulation: In this semester, we started with a literature survey of various implementations of the spatial modulation schemes. We found that mainly three variations of spatial modulation schemes: (i) Spatial Modulation (SM) (ii) Index Modulation (IM) (iii) Subcarrier Number Modulation (SNM), are widely studied and reported as the potential modulation schemes for next-generation wireless communication due to their attractive trade-off between the spectral efficiency and robustness, and better implementation complexity. Further, we observed the efficient applications of Deep Learning (DL) methods in image processing, computer vision problems, and recent studies in wireless communication have also demonstrated Deep learning’s potential for future generations of wireless networks. Therefore, we aim to develop a feasible wireless PHY layer for next-generation communication using the discussed technologies. Firstly, we implemented the spatial modulation (SM) with orthogonal frequency division multiplexing (SM-OFDM) system and generated results similar to original SM-OFDM work [7]. Further, we have combined SM-OFDM with a Neural Network (NN) or Convolutional Neural Network (CNN) based bit-error correction block that will improve the robustness (BER) of the proposed system without degrading its spectral efficiency. In other words, we can say that DL bit-error correction block at receiver will mitigate the need for channel coding; thereby, the proposed has the advantage of both throughput and robustness.