In January of 2017, we prototyped a design to test if the photo-diode and detection circuitry could provide the desired noise equivalent power and dynamic range. A photograph of the prototype device is shown in Fig. 1. The device contains two LEDs, emitting at 735 nm and 810 nm, with up to 8 mW optical output power. The LED power is supplied by their own voltage regulator to minimize contamination of the supply for the detection side. The LEDs can also be driven at a second, lower power level needed for short separation measurements. Light radiating from the tissue is collected by a photodiode with an integrated current amplifier. The integrated amplifier allows the device to be near shot-noise limited, and reduces the sensitivity to electro-magnetic interference. The output current of the photodiode module is sent through a trans-impedance amplifier (TIA), which serves as a low-pass and anti-aliasing filter, as well as a driver for the analog to digital converter (ADC). The ADC features 24-bit resolution and a conversion rate of up to 1 MSPS. It also contains a built-in digital averaging unit (32-fold), which greatly reduces required computation further down in the signal path. The datasheet performance of 98 dB dynamic range at 1 MSPS, or 128 dB at 1kSPS is more than enough for our application. Finally, the prototype optode also contains a small field-programmable gate array (FPGA) to fetch data from the ADC, perform further averaging (256-fold), and transmit the data to the data aggregation and timing control unit. In the future, this FPGA will also be used to modulate the LEDs at particular frequencies, and to perform the first steps of demodulation, similar to what we have previously described in . Currently, the signal is acquired at 983040 samples per second at the photodiode level, then averaged 32 times in the ADC and 256 times in the FPGA, for a resulting output data rate of 120 sps. This should dramatically reduce 120 Hz fluorescent tube flicker. In the data aggregation/control unit, we average an even number of samples, thus integrating out 60 Hz power line noise.
Performance testing of our prototype was performed by sending the light of one module through a filter wheel containing neutral density filters of known absorption. The transmitted light was then measured with a second optode module, acquiring and averaging 0.5 s of data with the LED turned on, and 0.5 s with the LED off. The “off” signal was subsequently subtracted from the “on” signal, and the resulting net signal plotted versus optical power reveals a noise equivalent power (NEP) of 0.5 pW / Hz1/2 with a >100dB dynamic range, meeting our design specification. This was achieved with a transimpedance of 100k. Increasing this will improve the sensitivity. We can reduce noise by adding more low pass filtering on the voltage reference. In addition, 1/f noise can be reduced by utilizing lock-in detection by modulating the light source at 1 kHz. With these additions, we are confident that we will achieve a 10x improvement in NEP, approaching the component specification of 0.01 pW / Hz1/2. As documented in , and consistent with our own experience, a 1 pW / Hz1/2 NEP permits measurements on the human head with source detector separations up to 60 mm, providing SNRs of 100 and 10 at 30 and 45 mm separations respectively. As our experience indicates, this provides ample SNR for measuring brain activity.
 B. B. Zimmermann, Q. Fang, D. A. Boas, and S. A. Carp, “Frequency domain near-infrared multiwavelength imager design using high-speed, direct analog-to-digital conversion,” J. Biomed. Opt., vol. 21, no. 1, p. 16010, 2016.
 D. Chitnis, R. J. Cooper, L. Dempsey, S. Powell, S. Quaggia, D. Highton, C. Elwell, J. C. Hebden, and N.L. Everdell, “Functional imaging of the human brain using a modular, fibre-less, high-density diffuse optical tomography system,” Biomed. Opt. Express, vol. 7, no. 10, p. 4275, 2016.