NeuroDAC: an open-source arbitrary biosignal waveform generator

Abstract

Objective.Researchers are developing biomedical devices with embedded closed-loop algorithms for providing advanced adaptive therapies. As these devices become more capable and algorithms become more complex, tasked with integrating and interpreting multi-channel, multi-modal electrophysiological signals, there is a need for flexible bench-top testing and prototyping. We present a methodology for leveraging off-the-shelf audio equipment to construct a biosignal waveform generator capable of streaming pre-recorded biosignals from a host computer. By re-playing known, well-characterized, but physiologically relevant real-world biosignals into a device under test, researchers can evaluate their systems without the need for expensivein vivoexperiments.Approach.An open-source design based on the proposed methodology is described and validated, the NeuroDAC. NeuroDAC allows for 8 independent channels of biosignal playback using a simple, custom designed attenuation and buffering circuit. Applications can communicate with the device over a USB interface using standard audio drivers. On-board analog amplitude adjustment is used to maximize the dynamic range for a given signal and can be independently tuned for each channel.Main results.Low noise component selection yields a no-signal noise floor of just 5.35 ± 0.063. NeuroDAC's frequency response is characterized with a high pass -3 dB rolloff at 0.57 Hz, and is capable of accurately reproducing a wide assortment of biosignals ranging from EMG, EEG, and ECG to extracellularly recorded neural activity. We also present an application example using the device to test embedded algorithms on a closed-loop neural modulation device, the Medtronic RC+S.Significance.By making the design of NeuroDAC open-source we aim to present an accessible tool for rapidly prototyping new biomedical devices and algorithms than can be easily modified based on individual testing needs.ClinicalTrials.gov Identifiers: NCT04281134, NCT03437928, NCT03582891.

Keywords: biomedical devices; biosignal playback; closed-loop neuromodulation; neural interface; waveform generator.

Creative Commons Attribution license.

Figures

Figure 1.

(A) An accessible low-cost consumer grade off-the-shelf audio DAC can be used in conjunction with simple analog attenuation and buffering circuitry to build a biosignal waveform generator. (B) NeuroDAC can be used to play back re-recorded electrophysiology measurements at physiological voltages to perform bench-top verification and validation testing of biomedical devices using real-life signals.

Figure 2.

(A) Schematic diagram of a single channel of the NeuroDAC device. A computer is used to stream data to the U-DAC8 via USB and a conditioning circuit converts the line-level voltage output to the small scale signals expected by bioelectronic devices. The conditioning circuit comprises 8 identical channels each containing an adjustable voltage divider circuit and a low noise op-amp based buffer with switches to independently configure the circuit as desired by the user. (B) An image of the NeuroDAC device depicting the U-DAC8 (miniDSP, Hong Kong) audio DAC and the custom-built attenuation circuit with BNC outputs for easily connecting to downstream devices. The U-DAC8 is powered through the conditioning board so only one power adapter is required.

Figure 3.

(A) A low voltage noise op-amp was chosen as the output buffer because this type of noise dominates in the operating regime of the circuit. A comparison is given between the voltage and current noise of the LT1007 op-amp and the Johnson noise of the voltage divider circuit for different settings of the potentiometer (switched resistor at 1 MΩ) at 1 kHz. The shaded region represents the range of resistances available for the 2 kΩ potentiometer used in NeuroDAC. Theoretical total noise accounting for all of these sources is also shown at 10 Hz and 1 kHz. (B) The full theoretical voltage noise spectrum of the LT1007 op-amp including low frequency flicker noise. (C) The total estimated noise spectra of the NeuroDAC analog circuitry at potentiometer settings of 100 Ω, 1 kΩ, and 2 kΩ. Cumulative RMS noise is indicated in blue as the bandwidth of the system is selected from 0.1 Hz to the value indicated on the horizontal axis. The shaded blue region indicates the integral area underneath the noise density curve for the worst-case 2 kΩ potentiometer setting. A dashed line indicates the reported noise floor of the Intan RHD2132 biosignal amplifier for reference (2.4 μVrms [41]).

Figure 4.

(A) A raw 1 second example trace of the Intan RHD2132 amplifier and OpenEphys data acquisition system noise floor [42]. Mean RMS noise floor (± std.) for all 5, 10 second long, recordings is overlayed as a dotted black line. (B) A raw 1 second example trace of the NeuroDAC noise floor recorded through the Intan/OpenEphys acquisition system. Mean RMS noise floor (± std.) for all 5, 10 second long, recordings is overlayed as a dotted black line. (C) A comparison of the estimated noise floor density between the data acquisition system with and without the NeuroDAC attached. The theoretical estimation of the noise density is given as a black dotted line. An expanded view of the frequency range from 0 to 250 Hz (blue highlighted region) is shown on the right.

Figure 5.

(A) Mean recorded DAC output for 10 repetitions of a linear, full-scale, sweep of input values. Black dotted line is the best-fit-line of the result found using least squares linear regression (R2 value reported in boxed annotation). (B) Per sample residuals for the waveform shown in (A). Values normalized to the slope of the best-fit-line.

Figure 6.

(A) Magnitude of the frequency response of the NeuroDAC system. Amplitude is attenuated for signal components lower than 1 Hz, but are accurately represented up to 10 kHz. Sub-figures call out individual data points used in the estimation, showing the original signal, raw playback signal, and model fit used to estimate amplitude parameter. (B) Phase of the frequency response of the NeuroDAC system. A phase shift is observed at the extrema of the operating bandwidth, however prediction accuracy may be decreased at higher frequencies due to the low sample/period ratio of the 30 kSPS data acquisition system. Sub-figure callouts indicate individual data points used to calculate phase shift.

Figure 7.

Measured output voltage of NeuroDAC while playing a 100 Hz, 250 μV sine wave into saline. The output was measured with and without buffering outside of the saline (‘Wired’), loaded by the saline (‘Loaded’), and through the saline using a tungsten electrode (‘Saline’). (A) A 0.1 s window of the raw recorded data from each condition. The waveforms were phase aligned in post processing to allow for easy comparison of amplitudes. (B) The estimated amplitude of the recorded waveform in each condition; calculated from 10 s of recording. Error bars indicate the 95% confidence interval of the estimated amplitude.

Figure 8.

A set of 6 different calibrated biosignals played through the NeuroDAC, overlayed and aligned to the original signal. In each sub-figure, a subset of the 30 second recording is shown on the left and an expanded view of the highlighted blue region is shown on the right highlighting a feature of interest for each signal. Different time and voltage scales are used due to the heterogeneity of signal characteristics. The signals represented are: (A) a microelectrode array recording of action potentials from a nonhuman primate; a single unit action potential is highlighted, (B) scalp-EEG recorded from a human subject; an alpha oscillation is highlighted, (C) intraoperative intracranial recording from a human subject; 130 Hz stimulation artifact is highlighted, (D) surface EMG activity recorded from an awake sheep; evoked activity from spinal cord stimulation is highlighted, (E) ECG recorded from a human subject; a single waveform is highlighted, and (F) EOG activity recorded from a human subject; a single blink event is highlighted. For more information about each signal, please refer to the main text body.

Figure 9.

Examples of using NeuroDAC to perform bench-top testing of an adaptive neuromodulation device depicting two use cases: (A) an artificial LFP burst sequence at nominal frequency of 20 Hz and varying amplitude between 50 and 10 μV and (B) playback of real neural data from GPi of a DBS patient. The panels show (from top to bottom), the time domain signal played by NeuroDAC and recorded by the RC+S, a spectrogram (0–50 Hz) of the signal, the output of the RC+S’s linear discriminant detector for in-band signal power, the stimulation current delivered by the RC+S, and the embedded algorithm’s decoded state for adaptive stimulation.