A pulse oximeter based on Time-of-Flight histograms

A pulse oximeter is an optical device that monitors tissue oxygenation levels. Traditionally, these devices estimate the oxygenation level by measuring the intensity of the transmitted light through the tissue and are embedded into everyday devices such as smartphones and smartwatches. However, these sensors require prior information and are susceptible to unwanted changes in the intensity, including ambient light, skin tone, and motion artefacts. Previous experiments have shown the potential of Time-of-Flight (ToF) techniques in measurements of tissue hemodynamics. Our proposed technology uses histograms of photon flight paths within the tissue to obtain tissue oxygenation, regardless of the changes in the intensity of the source. Our device is based on a 45ps time-to-digital converter (TDC) which is implemented in a Xilinx Zynq UltraScale+ field programmable gate array (FPGA), a CMOS Single Photon Avalanche Diode (SPAD) detector, and a low-cost compact laser source. All these components including the SPAD detector are manufactured using the latest commercially available technology, which leads to increased linearity, accuracy, and stability for ToF measurements. This proof-of-concept system is approximately 10cmx8cmx5cm in size, with a high potential for shrinkage through further system development and component integration. We demonstrate preliminary results of ToF pulse measurements and report the engineering details, trade-offs, and challenges of this design. We discuss the potential for mass adoption of ToF based pulse oximeters in everyday devices such as smartphones and wearables.

💡 Research Summary

This paper presents a novel pulse oximetry approach that leverages Time‑of‑Flight (ToF) histograms rather than traditional intensity measurements to estimate blood oxygen saturation. The authors built a proof‑of‑concept system comprising a 45 ps resolution Time‑to‑Digital Converter (TDC) implemented in a Xilinx Zynq UltraScale+ FPGA, a CMOS Single‑Photon Avalanche Diode (SPAD) detector, and two laser sources: a commercial Hamamatsu pulsed laser and a low‑cost custom‑developed laser. The SPAD is a single 8 µm pixel with a dark count rate of 1 kHz, a 20 ns recovery time, and photon detection probabilities of 25 % at 650 nm and 10 % at 773 nm. The TDC uses a tapped‑delay‑line (TDL) architecture with 448 delay taps, 37 histogram bins of 45 ps average width, and a measurement range of 1.66 ns. Code‑density testing shows differential non‑linearity (DNL) within ±0.15 LSB and integral non‑linearity (INL) within ±0.55 LSB, with no missing codes, confirming high linearity essential for accurate ToF measurements.

Data processing extracts the histogram bin counts, normalizes them, and computes the mean flight time ⟨t⟩ via a Mellin‑Laplace‑based second‑order transform. ⟨t⟩ reflects the shape of the temporal spread function (TPSF) and is theoretically independent of the total photon count (i.e., light intensity). This property is central to the proposed method: changes in tissue absorption (due to varying oxy‑ and deoxy‑hemoglobin concentrations) alter the TPSF shape and thus ⟨t⟩, while fluctuations in source power, ambient light, or skin tone affect only the overall photon count, leaving ⟨t⟩ essentially unchanged.

The experimental program includes three major tests. First, a stability test with the Hamamatsu laser collected 45 k frames at 50 fps over 15 minutes. While the raw intensity varied by 41 %, the mean flight time drifted by only 3 %, demonstrating robustness of ⟨t⟩ against intensity drift. Second, neutral‑density (ND) filters (Thorlabs NE10A, NE20A‑B, NENIR40A‑C) were inserted to induce a 138 % change in intensity; ⟨t⟩ changed by less than 1 %, confirming intensity independence. The filter trials also revealed increased noise when photon counts dropped, as expected, but the average ⟨t⟩ remained stable. Third, a Vascular Occlusion Test (VOT) was performed by inflating a cuff on the upper arm to 160 mmHg for ~30 s and then releasing it. Both the commercial and low‑cost lasers recorded a clear decrease in ⟨t> during occlusion (higher total absorption yields a narrower TPSF) and a recovery after release. Importantly, when an ND filter was applied only during the initial baseline, intensity dropped sharply while ⟨t> trend stayed unchanged, reinforcing the decoupling of ⟨t> from intensity.

System dimensions are currently 10 cm × 8 cm × 5 cm, with the low‑cost laser module measuring 4 cm × 5 cm × 0.5 cm. The authors argue that integrating the FPGA, SPAD, and laser onto a single System‑on‑Chip (SoC) could shrink the device to a form factor suitable for smartphones or smartwatches.

Limitations are acknowledged. The prototype uses a single wavelength (650 nm); full oximetry requires at least two wavelengths (e.g., 660 nm for oxy‑hemoglobin and 940 nm for deoxy‑hemoglobin) to resolve the ratio of the two chromophores. The low‑cost laser does not fully turn off between pulses, resulting in a higher baseline and reduced dynamic range compared with the Hamamatsu source. Future work should focus on (i) adding a second wavelength and calibrating the relationship between ⟨t> differences and actual SpO₂ values, (ii) optimizing the custom laser driver to achieve sub‑20 ps optical pulses and true off‑states, (iii) implementing on‑FPGA real‑time histogram accumulation and ⟨t> computation to eliminate the need for PC post‑processing, and (iv) developing low‑power wireless data links for wearable integration.

In summary, the study demonstrates that mean photon flight time extracted from ToF histograms provides a metric that is intrinsically immune to source intensity variations, ambient light, and skin tone, addressing a major source of error in conventional pulse oximeters. By combining a high‑resolution FPGA‑based TDC with a low‑jitter SPAD detector, the authors achieve sub‑50 ps temporal resolution, enabling precise detection of hemodynamic changes. The approach shows promise for mass‑market adoption, offering a pathway to compact, low‑cost, and robust pulse oximetry that could be embedded directly into consumer electronics.