Abstract

Abstract--An integrated framework for non-contact respiratory
rate (RR) estimate using a Zynq-7000 SoC platform and
a 77 GHz FMCW radar is presented in this work. For shortrange
respiratory monitoring, three lightweight signal processing
techniques multitaper spectrum analysis (MSA), adaptive harmonic
component analysis (AHCA), and autoregressive linear
predictive coding (AR-LPC) are compared. AR-LPC achieves
the lowest estimation error among the evaluated approaches,
according to an experimental evaluation conducted in controlled
indoor settings, whereas MSA offers stable performance across a
range of distances. The viability of implementing computationally
light radar-based respiratory monitoring algorithms on portable
edge platforms is demonstrated by the suggested implementation.
In order to direct future development toward robust deployment,
limitations related to subject diversity, recording time, and
motion conditions are also explored.

Abstract

Radar sensor-based vital sign monitoring holds significant promise, but most of the existing approaches impose strict constraints, requiring subjects to be immobile and at a con- stant distance from the radar, which is less encountered in real-world scenarios. This preliminary study addresses these shortcomings by proposing a single lightweight 1D CNN net- work, inspired by ResNet, to accurately estimate the heart rate (HR) and the breathing rate (BR) with the 77 GHz mmWave FMCW radar sensor. The model is trained to estimate vital signs under motion scenarios with minimal pre-processing. The model directly processes raw or near-raw radar signals, demonstrating strong resilience to motion artefacts common in real-world settings. The proposed network achieves a mean absolute error (MAE) of 2.3 beats per minute for HR and 1.67 breaths per minute for BR. The model is efficiently deployed on a PYNQ-ZU edge platform with an inference time of just 22.6 milliseconds, showcasing its suitability for real-time, portable edge device-based health monitoring applications.

Abstract

This paper presents design of low-power, low-area, PVT-invariant 3-stage differential ring oscillator (RO) targeting a frequency of 1 GHz. The design uses a process-trimmed, temperature-compensated robust current reference, designed in TSMC 65 nm CMOS technology, achieving a temperature coefficient as low as 21.67 ppm/° C across temperatures from −40° C to 125° C. A detailed, quantitative simulation focussed design methodology and implemenation is proposed for the design of the current reference and mirroring circuit. The resulting oscillator shows less than 0.5% frequency variation across temperature and a worst-case variation of only 4.78% across PVT. To ensure sub-1V supply operation, only lowthreshold (LVT) MOSFETs are used throughout the design. The proposed design is a low power design and the oscillator consumes approximately 0.36 mW of power at 800 mV supply. This power efficient architecture is well-suited for PVT-robust on-chip clocking in modern low-voltage systems.

Abstract

This paper presents a low phase-noise LC-VCO for upper Ku-band operation using a Gm,eff-boosted core with a triple-coupled stacked transformer and noise-circulation. Fabricated in 65-nm CMOS, it achieves a 15-18.3 GHz (19.8%) tuning range, -114.8 dBc/Hz phase noise at 1 MHz offset, and FoM/FoMT/FoMA of 189.1/195.0/201.0 dBc/Hz. The VCO occupies
0.064 mm2, consumes less than 9 mW from 1 V, and is suited for compact upper-Ku/mmWave synthesizers.

Abstract

This paper presents a 13.95-16.96 GHz low power, low phase noise (PN) LC voltage-controlled oscillator (VCO) in 65 nm CMOS technology.
To address the limitations of conventional harmonic tuned oscillators, whose phase noise (PN) degrades significantly across the tuning range (TR) due to narrowband harmonic impedances, a custom transformer is designed and analyzed.
The proposed tank can actively contribute to ISF shaping and achieve consistent flicker noise upconversion reduction across the entire tuning range. Post-layout simulations demonstrate a phase noise (PN) of -118 dBc/Hz and a figure of merit (FoM) exceeding 187 dBc/Hz at a 1 MHz offset. The VCO occupies only 0.0306 mm$^2$, and its PN performance makes it suitable for mm-Wave applications and next-generation wireless systems.

Abstract

Reliable road object detection is crucial for self-driving cars and intelligent transportation systems. However, vision-only techniques fail in low-visibility conditions such as fog, rain, or nightfall, whereas deep radar-based models frequently suffer from high latency and energy consumption. To overcome these restrictions, we propose a millimeter-wave (mmWave) radar-based object identification system based on a PYNQ-ZU (Field-Programmable Gate Array) FPGA that achieves a compromise between accuracy, efficiency, and real-time performance. Our proposed pipeline transforms 3D radar point clouds to 2D Top-View (TV) representations and uses the Spectral Graph Wavelet Transform (SGWT) to extract discriminative spatial-temporal features with little computational cost. A Random Forest (RF) classifier achieves 97% accuracy across seven object classes with an end-to-end latency of 61 ms. When it comes to power performance, the FPGA implementation uses 119.32 mW at idle, 268.48 mW on average, and 357.97 mW at its peak, demonstrating its energy efficiency under dynamic workloads. Compared with deep learning alternatives like Tiny Convolutional Neural Network (CNN) (73.60 MB, 4493 mW, 449 ms), the model size is 6.47 MB, which is substantially smaller and more power-efficient. By overcoming the limitations of vision-only systems (poor visibility) and deep radar models (high latency, energy, and memory), the proposed SGWT + RF framework enables real-time, multi-class detection in resource-constrained environments--offering a practical and robust solution for autonomous navigation and traffic monitoring applications.

Abstract

This paper presents a constant gm, 1.2 V operational transconductance amplifier (OTA) designed in TSMC 65 nm CMOS technology. The proposed OTA achieves a constantgm characteristic over a full rail-to-rail input/output range by employing a constant gm rail-to-rail input stage, a folded cascode summing stage, and a quiescent current controlled Class-AB output stage. The input stage comprises complementary NMOS and PMOS differential pairs connected in parallel to achieve rail-to-rail input operation. Conventional parallel input stages suffer from significant gm variation with input common-mode voltage, leading to changes in small and large signal parameters such as open-loop gain, slew rate and unity-gain bandwidth (UGB). In this work, a current-injection mechanism dynamically adjusts the active input pair's bias, maintaining a nearly constant transconductance across the entire common-mode range. Postlayout simulation shows that the proposed OTA achieves less than 4.8% gm variation across the entire input range while maintaining uniform large signal characteristics. Performance metrics include a DC gain of 80-87 dB, UGB of 9.72 MHz, phase margin of 64°, CMRR of 97.9 dB, and a power consumption of 260 μW when driving a 100 pF capacitive load.

Abstract

This work proposes a practical and privacy-preserving gesture recognition system using millimeter-wave (mmWave) radar sensors for low-power, real-time edge computing applications. The solution uses a Top-View (TV) Filter to transform 3D point clouds into 2D point cloud images, providing a compact and informative representation of hand gestures while avoiding privacy concerns inherent to optical systems. Machine learning models are applied to classify 7-12 distinct gestures. Evaluation is performed on a publicly available dataset and a proprietary mmWave radar dataset, demonstrating high accuracy: Random Forest (RF) achieves 97% and 89% for 7 and 12 gestures, respectively, while Logistic Regression(LR) attains 79% and 78% for 10 and 9 gestures. Hardware validation on a PYNQ-ZU FPGA confirms low power consumption (3.2 mW) and minimal latency (0.67 ms), demonstrating the solution's suitability for practical, real-time deployment.

Abstract

Autonomous vehicles must accurately identify a variety of road objects, including cars, pedestrians, and bicycles, in order to operate safely in inclement weather and limited visibility. Millimeter-wave (mmWave) radar is consistently successful in these situations, although LiDAR and camera systems frequently fail. Despite this benefit, the existing radar-based detection systems have limitations: they are too computationally costly for real-time use, frequently limited to single-class identification, and perform poorly with the sparse, irregular nature of radar point clouds when using the standard grid-based convolutional neural networks (CNNs). To address the prevalent issue of high computational demands that limit embedded system implementation, we presented a real-time, low-latency framework for multiclass object classification utilizing mmWave radar. Our method uses 2-D top-view (TV) and front-view (FV) representations of radar point clouds, with the TV projection outperforming the FV projection in terms of spatial organization. To extract features, we used the graph Fourier transform (GFT), which uses the graph Laplacian to transform irregular input into a spectral domain while effectively capturing both macrostructures and microdetails. This approach, together with lightweight classifiers, yielded exceptional results: random forest (RF) and logistic regression (LR) reached 99.51% and 99.15% accuracy. LR also demonstrated an inference latency of just 15 ms/frame. The entire system was implemented on a PYNQ-ZU FPGA, demonstrating its scalability, efficiency, and real-time performance for robust, multiclass road object categorization in self-driving vehicles.

Abstract

This paper presents the design of a programmable Linear Frequency Modulator (LFM) which can be integrated with a phase-locked loop (PLL) based frequency synthesizer for edge AIoT applications such as radar sensing and wireless communication, and demonstrates its implementation on an FPGA. The LFM operates dynamically by adjusting the dividing value of the multi-modulus divider (MMD) within the feedback loop of PLL, which enables precise, stepped frequency sweeps and frequency hopping with a better digital control over step size, duration, and slope of the generated chirp. This programmability allows the system to adapt to varying application requirements, which offers enhanced flexibility, power efficiency, and timing precision. This LFM has a deterministic on-chip performance with an estimated power consumption of less than 108mW, making it suitable for Edge devices that are constrained by power and area. The proposed approach supports flexible frequency synthesis that improves communication reliability, interference resilience, and multi-functional sensing capabilities. This work highlights the critical role of programmable PLL-based frequency modulators in advancing compact, energy-efficient, and intelligent edge AIoT infrastructures across diverse domains.