Intsy: a low-cost, open-source, wireless multi-channel bioamplifier system

The bioamplifier system encompasses both a novel electronics module as well as user-friendly front-end software to interface with the hardware. Design decisions were guided by a 'Maker' philosophy (Anderson 2012, Hatch 2013) that the system be capable of high-quality recordings whilst being inexpensive, easy to assemble, and open-source so that others may build and customize their own modules. Therefore, we opted to use state-of-the-art commercial off-the-shelf components, which reduces the time and monetary expense inherent with development of custom ASICs, while also making the module much easier to replicate by others. Using commercial components leverages recent rapid advances in significantly increased hardware capability, on-board integration, and decreased physical footprint.

As illustrated in figure 1, the hardware module integrates four main components on a simple custom PCB: (1) 2  ×  Intan RHD2132 32-channel amplifier board (Intan Technologies; Los Angeles, CA); (2) Teensy 3.2 microcontroller (PJRC; Sherwood, OR); (3) Texas Instruments SN65LVDT41 low-voltage differential signaling (LVDS) line driver/receiver; and (4) RN-42 bluetooth modem (Sparkfun; Niwot, CO). The Intan- and Teensy- based hardware system is coined 'Intsy'.

Zoom In Zoom Out Reset image size

Each RHD2132 amplifier board offers 32 unipolar inputs with all analog and digital components—filters, amplifiers, multiplexer, A/D converter, and serial-peripheral interface (SPI)—fully integrated in a single, small footprint (24 mm  ×  15.5 mm) device (Intan Technologies 2013). Two 32-channel amplifier boards are used for the 64-channel system; only one is required for the 32-channel configuration. The Intan chip's specifications for low input-referred noise level (2.4 μV) and dynamic range (±5 mV; resolution of 0.195 μV/bit) are well-suited for measuring GI bioelectric signals, with typical amplitude ranges of  ≈0.5–2 mV for gastric serosal SWs (Egbuji et al 2010), ≈50–200 μV for the HR-EGG (Gharibans et al 2017), and surface EMG signals which typically range from about 100–1000 μV. An external 100 MΩ resistor was added to the amplifier board such that the on-board analog filter lower cut-off could be set to 0.02 Hz, beneficial for measuring gastric slow waves with typical frequency of  ≈0.05 Hz (3 cycles per minute) (Lin et al 2000).

We selected the Teensy 3.2 microcontroller for our system because it offers a relatively fast 72 MHz processor, an easy to program Arduino-like environment with mature ecosystem, a small form factor, on-board 3.3 V, 250 mA low-dropout (LDO) regulator to power peripherals, and full speed (12 Mbps) USB communication with the host computer. The Teensy communicates with (up to) two RHD2132 chips via a standard four-wire serial-peripheral interface (SPI) using LVDS. We used the TI SN65LVDT41 to translate between the native CMOS interface on the Teensy and LVDS signaling on the Intan chip. Each RHD2132 chip SPI-MISO line was programmed to be in high impedance state when the chip was not selected; this allows multiple chips to share most SPI lines, allowing a more compact design.

The Teensy was programmed to interactively configure the Intan chips' band-pass filter cutoff frequencies, and to control each 16-bit A/D conversion of analog channel input signals. The start of each sampling sequence was triggered by hardware timer interrupts with user-selectable timing intervals to precisely control the sampling period. All 32 or 64 input signals were sampled sequentially in round-robin fashion; all inputs are multiplexed through a single analog-to-digital converter (ADC) on board the RHD2132. The SPI clock was set to run at 12 MHz. In order to synchronize to external hardware, the Intsy system outputs a 3.3 V digital pulse at the start and stop of data acquisition.

The host computer (end-user) communicates with the Teensy using either a full-speed (12 Mbps) wired USB or Bluetooth 2.1  +  EDR serial connection. In the latter case, a Microchip RN-42 Bluetooth module with level shifting circuitry was used (Sparkfun Electronics Bluetooth Mate Silver) set for 115 200 baud. CTS/RTS flow-control (MDM-SPP mode) was also enabled to achieve robust data streaming. We selected the RN-42 because it was easy to program and achieved stable data streaming much more reliably than alternative modules (e.g. the JY-MCU HC-06). In either wired or wireless configuration, 32 or 64 channels of analog data was transmitted along with a 4-byte timestamp plus 4 extra bytes used to verify SPI connection fidelity with the RHD2132 chip. Intsy data transmission bandwidth was measured using a host computer with Intel core i7, 3.4 GHz processor running Windows 7^TM operating system.

All components were connected on a custom PCB with a 9  ×  5 cm footprint, or about the size of a credit card. Schematic design and PCB layout was done in Eagle 7.7.0 software and manufactured by Seeed Studio Fusion. All further assembly was done in-house. Module assembly typically required 2–4 h, and can be performed with basic electronics assembly skills. In order to make assembly as simple as possible, we opted to use through-hole components, where appropriate. The LVDS line driver was mounted on a 0.65 mm SOIC to 0.1" DIP adapter (Schmartboard, 204-0006-01). In addition, we integrated Intan's SPI cable adapters into the design, which interface between a surface-mount Omnetics 'nano' connector to a conventional 0.1" pitch through-hole header. Intan RHD2132 amplifier boards were connected either with a 12-contact SPI interface cable, or directly to the SPI cable adapter, as we opted to do during validation studies (see sections 2.5 and 2.7).

For wireless operation, the Insty hardware module is powered by a standard 3.7 V LiPo battery through a polarity-protection Schottky diode. For USB-wired mode, the host-computer USB port powers the Intsy module through a commercially available USB isolator with isolated DC–DC power supply (Adafruit, part number 2107), modified to be compatible with 3.3 V logic. In either case, power is routed through the Teensy's on-board 3.3 V LDO voltage regulator which distributes power to all components on the PCB. The total current budget for wirelessly streaming data was 113 mA (Teensy 3.2  ≈  27 mA; Intan RHD2132  ≈1 mA; LVDT41  ⩽35 mA; Bluetooth RN-42  ⩽50 mA). Thus, a standard 1000 mAh LiPo battery would be expected to power the Intsy hardware module running in wireless mode for about 8 h.

The Teensy 3.2 was programmed using Teensyduino software, an add-on to the standard Arduino integrated development environment (version 1.8.3).

A custom, user-friendly software library for communicating with the Intsy hardware module was written in Labview 2015. LabView was chosen because it is already in use at many academic institutions and research labs, and has an extensive library for serial port read and write protocols. The LabView graphical user interface allows the user to select wired or wireless data transfer, configure Intan filter cutoffs, set the sampling rate, visualize signals in real-time, and stream data to hard-drive for further off-line analysis. Real-time visualization features include software-filtering signals, selecting a subset of channels to display, and setting the amplitude and time-scales appropriate for the incoming data.

To demonstrate practical utility of the Intsy system applied to various bioelectric domains, the device performance was validated with two in vivo studies measuring (i) serosal slow wave activity in the porcine stomach, and (ii) human forearm muscle contractions with cutaneous adhesive electrodes, i.e. the surface electromyogram (sEMG). Performance of the Intsy module was compared to a gold-standard#157; commercial system, the Biosemi ActiveTwo (Biosemi, Amsterdam, Netherlands), modified for passive recordings (O'Grady et al 2012, Angeli et al 2015, 2017).

A gastric slow wave validation study was conducted to demonstrate the applicability and accuracy of the Intsy device for gastrointestinal recordings. Ethical approval for this study was granted by the University of Auckland Animal Ethics Committee. Serosal recordings were collected and analyzed from six pigs. Animal preparation and surgical procedure were as previously validated and described (Egbuji et al 2010). Serosal recordings were performed using tessellated flexible-printed-circuit (FPC) arrays ((Du et al 2009); FlexiMap, Auckland, New Zealand) with 256 total electrode contacts (8  ×  32 configuration; 4 mm spacing; 35 cm²). The electrode coverage extended from the mid-corpus to pylorus (figure 2(b)). 32 electrodes were routed to the Intsy system wirelessly streaming the data to the laptop computer, and the remaining 224 electrodes were routed to a Biosemi ActiveTwo system. Reference electrodes for each recording device were placed on the hindquarter thigh. Signals were recorded at a sampling rate of 30 Hz from the Intsy system configured for wireless data transmission. Signals were recorded at a sampling rate of 512 Hz from the Biosemi system, which were subsequently downsampled to 30 Hz during off-line analysis.

Zoom In Zoom Out Reset image size

All signals recorded by the Intsy and the Biosemi systems were processed using a previously validated analysis pipeline, implemented in the GEMS v2.2 software package (Yassi et al 2012). Signals were filtered to reduce baseline wander (moving median filter, 10 s) and high frequency noise (Savitzky–Golay filter, order 9, 1.7 s) (Paskaranandavadivel et al 2011). Slow wave activation times (ATs) were marked with the automated FEVT algorithm (Erickson et al 2010), which were then used to produce isochronal activation and velocity field maps (Paskaranandavadivel et al 2012, Yassi et al 2012). Slow wave intervals were calculated as the time difference between two successive ATs.

To further compare SW waveforms recorded with each system, the signal to noise ratio (SNR) was computed as

$$\begin{align} \newcommand{\e}{{\rm e}} \displaystyle \label{eqn:SWsnr} {\rm SNR_{SW}\;({\rm dB})}~=~20\log_{10}\left(\frac{V_{pp}}{\hat{\sigma}_{b}}\right) \nonumber \end{align} \tag{ 1 }$$

where V_pp indicates the peak-to-peak amplitude within a 20 s window centered on the marked AT (i.e. the main downstroke), and $\hat{\sigma}_{b}$ indicates the baseline noise estimate, which was measured from two 1.5 s windows at the beginning and end of the 20 s window (e.g. 0–1.5 s and 18.5–20 s). These noise windows worked in practice because slow wave intervals were  ≈20 s and the activation-recovery intervals are typically  ⩽7 s (Paskaranandavadivel et al 2017b). The peak-to-peak amplitude was used here because it does not assume a specific stereotyped morphology.

The Mann–Whitney U test was used for statistical comparison of the median value of SW interval and SNR distributions, with a significance threshold based on a 95% confidence level; $p\geqslant 0.05$ indicated median values were statistically indistinguishable. Data were pooled across all 32 channels of the Intsy device and the subset of 32 immediately adjacent Biosemi channels (e.g. the electrode sites directly above and below the dashed box in figure 2(c)). Median statistics were used to avoid the influence of false positive SW marks, which would lead to outliers in the distribution of SW intervals and/or SNR. Quantitative metrics are reported as a mean  ±  s.d., unless otherwise specified.

A sEMG validation study was conducted to demonstrate applicability of the Intsy device to a broader range of biomedical applications. Given the non-invasive nature and ease of application, the sEMG has been extensively utilized for biomechanical studies such as activation timing (De Luca 1997), biofeedback rehabilitation therapy for musculoskeletal conditions (Giggins et al 2013), and non-invasive control of prosthetic limbs (Castellini and van der Smagt 2009). One robust control strategy still commonly implemented in upper limb prostheses is level coding, actuating a desired prosthetic function based on the strength of muscle activity relative to maximal contraction (Farina et al 2014). Therefore, our validation study consisted of participants repeatedly squeezing a custom hand-held torsional spring device between the fingers and palm with 4 cm lever arm, essentially making a fist. Participants cycled through five prescribed force levels ranging from 6.4–14.2 N spaced at equal intervals.

Ethical approval for this study was granted by The University of Auckland Human Participants Ethics Committee; informed consent was obtained from each participant. Following skin preparation with alcohol and NuPrep, three Ag/AgCl adhesive cutaneous electrodes (22  ×  22 mm, Nikomed 2001) were affixed to the forearm above the belly of the flexor carpi radialis muscle to record the sEMG. Additionally, two adhesive electrodes were affixed to the medial epicondyles of the elbow served for ground and reference. Snap-lead cables were affixed to the forearm with medical tape to minimize motion artifacts.

Signals were recorded with the Intsy module configured for wireless (Bluetooth) and wired (USB) data transmission at sampling frequencies of 100 Hz and 1 kHz, respectively, and with with the Biosemi system set for 512 Hz sampling. The Intsy sampling frequencies were chosen because 100 Hz is near the maximum stable Bluetooth 2.1 data transmission bandwidth, and 1 kHz would be expected to properly measure components spanning the sEMG power spectrum, with dominant power typically concentrated in the range of  ≈20–200 Hz, and only a small fraction of signal power expected to occur at  ⩾250 Hz (Basmajian and De Luca 1985, Potvin 1997, Rojas-Martínez et al 2012). For each measurement mode, the 5-force level grasping task was repeated 10–12 times, with a 2 min rest period between each trial to avoid muscle fatigue. All trials were conducted in a typical office environment to mimic real-life conditions; no purposeful effort was made to minimize electromagnetic interference.

The sEMG signals were processed off-line using custom software written in MATLAB. Digital filters were applied to reduce low frequency components associated with movement artifacts (Butterworth, 2nd order, zero phase, high pass, 10 Hz cutoff) (De Luca et al 2010), and to reduce 50 Hz power line noise (Butterworth, 2nd order, stop band cutoffs of 48–52 Hz for 512 and 1000 Hz sampling rates, and single-sided 48–50 Hz filter for 100 Hz sampling rate).

The time-averaged RMS envelope was computed as a measure of the sEMG signal strength (power), which has long-standing use in sEMG signal analysis (De Luca 1997). Time averaging was done with a box-car filter of half-width T_w:

$$\begin{align} \newcommand{\e}{{\rm e}} \displaystyle {\rm RMS}[s(t_n)]~=~\frac{1}{N_w}\sum^{t_n + T_w}_{t_n - T_w} [s(t)]^{2} \nonumber \end{align} \tag{ 2 }$$

where s(t_n) is the sEMG signal recorded at integer time index t_n, and N_w is the number of samples within the box car window, dependent on the native sampling frequency. A value of T_w  =  50 ms was used, as this permits accurate automated estimation of onset activation times (Hodges and Bui 1996). Subsequently, sEMG spike burst (de-)activation times were manually marked by expert reviewers simultaneously viewing the sEMG and RMS signals.

In order to quantitatively compare Intsy and Biosemi system signal quality, as well as demonstrate practical utility of the new device, the SNR of the smoothed RMS signal during activation was computed as follows:

$$\begin{align} \newcommand{\e}{{\rm e}} \displaystyle \label{eqn:emgsnr} {\rm SNR_{EMG}\;(dB)}~=~20\log_{10}\left(\frac{A_{s} - A_{n}}{\hat{\sigma}_{n}}\right) \nonumber \end{align} \tag{ 3 }$$

where A_s represents the median of the smoothed RMS signal within the marked spike burst window (figure 5). In equation (3), A_n is the median of the RMS signal during baseline (muscle inactivated), and $\hat{\sigma}_{n}$ is the median of the absolute deviation,of the RMS during baseline, a robust estimator of the standard deviation (Nenadic and Burdick 2005). Median statistics were used in order to avoid the affect of brief and large amplitude components in the RMS signal, which can occur during the initial activation epoch.

In essence, the SNR quantifies the ease and accuracy with which spike burst on and off times can be detected. Note that a SNR  =  9.54 dB means that the EMG signal burst amplitude was 3  ×  above the baseline noise level. As a rule of thumb, signals with SNR  ⩾10 dB can be considered high quality, and even signals with SNR  ⩾3 dB are sufficient for detection of on and off times (Staude et al 2001).

The input-referred system voltage noise was measured by shorting all inputs to ground at the electrode connector contact points of the Intan chip, and computing the root mean square of each recorded signal over a 60 s window. The baseline system noise and interference levels in a typical biological preparation were assessed using epochs from sEMG recordings during which the forearm was relaxed (per section 2.7). In order to quantify the contributions of random noise versus power line interference sources, baseline recordings ($n=3$) were made in the wireless configuration with the distance of the Intsy module and test subjet set to d  =  0.5, 2, or 5 m from a known interference source (mains-powered computer work station.)

Analog-to-digital (A/D) conversion speed was quantified as the amount of time required to sequentially convert 32 input channels via SPI commands, averaged over 5 min of data acquisition.

Input-referred voltage noise levels were measured to be 2.8  ±  0.9 μV when configured for wireless operation, and 3.1  ±  0.3 μV for USB-wired mode. The different power sources (LiPo battery versus DC–DC converter plugged into USB port) likely accounted for the small difference. The total baseline noise plus interference level was $6.3~\pm~1.5$ μV far from the interference source (d  =  5 m), $7.7~\pm~1.2$ μV at an intermediate distance (d  =  2 m), and $24.2 \pm 12.7$ near the interference source ($d~=~0.5$ m). Therefore, random noise contributed  ≈6 μV to the baseline signal recording, while power line interference sources contributed up to  ≈18 μV. In practice, total baseline levels (noise  +  interference) of  ⩽10 μV were readily achievable, yielding high quality recordings in which bioelectric events were readily identified by manual and automated means.

Hardware-timed interrupts defining the beginning of each round-robin sampling sequence were always accurate to within  ⩽2 μs. The time required for sequential A/D conversion of 32 input signals plus two auxiliary SPI commands was 70  ±  2 μs. This equates to an interchannel sampling delay of 2.06 μs, and a theoretical maximum sample rate of  ≈400 kS s⁻¹. The actual A/D conversion time was longer than the 51 μs theoretical minimum, which may reflect the extra clock cycles required by the Teensy 3.2's CPU to toggle the state of the SPI chip select (CS) line.

Bluetooth wireless data transmission rates up to 110 Hz/channel  ×  32 channels (or 55 Hz  ×  64 channels) remained stable for the maximum test period of 1 h. USB wired data transmission remained stable up to the maximum tested rate of 2000 Hz/channel  ×  32 channels (1000 Hz  ×  64 channels). The maximum achieved wireless data transmission was sufficiently fast to make high-fidelity recordings of GI slow waves (section 3.2) and to clearly identify spike bursts in the sEMG section 3.3. However, the higher bandwidths achieved with the USB wired connection would be required to permit accurate measurement of bioelectrical events with faster temporal scales, such as gastrointestinal spikes (Lammers et al 2003), and to properly determine the full sEMG frequency spectrum.

The Intsy device recorded high fidelity signals with SW waveforms that were qualitatively indistinguishable from those recorded with the Biosemi system. Typical electrograms recorded with both systems are compared in figure 2(a). The main downstroke component associated with SW activation was readily identified by the automated FEVT algorithm. The resulting sub-region of the isochronal and velocity field maps derived from the Intsy SW marks were consistent with the surrounding regions derived from Biosemi recordings.

Subtle morphological differences were observed between slow waves recorded with the two systems (figure 3). The Biosemi typically recorded a modestly larger initial upstroke component associated with an approaching wavefront, whereas the Intsy typically emphasized the positive deflection of the repolarization upstroke associated with the passing wavefront before settling to baseline.

Zoom In Zoom Out Reset image size

Only modest differences were observed between the distribution of SW intervals determined from each measurement system, during both normal (Experiments 1–3, 5–6) and dysrhythmic (Experiment 4) SW propagation sequences (figure 4(a)). Mann–Whitney U test results indicated SW interval distributions were statistically similar in all but the fifth experiment; p-values for the six experiments were 0.29, 0.79, 0.80, 0.40, 0.002, and 0.60. The discrepancy for Experiment 5 was attributed to false-positive marks due to a relatively lower SNR.

Zoom In Zoom Out Reset image size

The median SNR of the SW recorded with the Intsy system was 24  ±  2.7 dB across all test subjects (figure 4(b)). There was only a slight difference in SNR of the Intsy system relative to the Biosemi (−0.15  ±  3.1 dB; median  =  0.68 dB; range  =  −4.1 to  +3.4 dB). Statistical results indicated significant differences in the median value of the SNR distributions for all test subjects, except for Experiment 3. However, this difference has no practical consequence because the SNR for both systems is sufficiently high for accurate automated SW marking (Erickson et al 2010).

In practice, the isochrone activation and velocity field maps derived from each hardware system were essentially the same in all cases, as false positive marks are rejected from further consideration during the wave front clustering stage in the analysis pipeline (Erickson et al 2011). Overall, the Intsy system recorded electrograms that consistently led to accurate mapping of SW propagation patterns during normal and dysrhythmic events.

Figure 5 shows a representative sEMG trace acquired with the Intsy system wirelessly steaming data at 100 Hz. Spike burst activation with increasing signal strength corresponding to five successively stronger forearm contractions is apparent to the naked eye. SNR values ranged from 10.2  ±  3.5 dB for weakest contractions to 19.4  ±  3.7 dB for the strongest (figure 6(a), red trace). This SNR range is sufficiently high in practice to precisely determine activation timing (Staude et al 2001) and patterns with machine learning algorithms, thus validating potential utility of this device for ambulatory sEMG applications, such as myoelectric control strategies with upper-limb prosthetics (Castellini and van der Smagt 2009, Farina et al 2014).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 6(a) compares the SNR versus contraction force levels for all three measurement modes. The expected positive correlation between increasing signal strength and contraction force (De Luca 1997) was readily apparent across all five test subjects for each mode (figures 6(b)–(d)). The observed intersubject variation can be attributed to differences in intrinsic physiology and extrinsic factors, such as electrode/muscle fiber orientation (De Luca 1997).

On average, the spike burst SNR measured with the Intsy system configured for 1 kHz sampling was similar to that of the Biosemi, with a modest average difference of 1.1  ±  0.2 dB. The SNR for the Biosemi system may have been underestimated in our experiment by  ≈1.0–1.5 dB due to aliasing  ≈15% of the sEMG power that may have occurred above the 256 Hz Nyquist frequency (De Luca 1997, Potvin 1997, Rojas-Martínez et al 2012). In practice, the Intsy's SNR is sufficiently high for accurate automated detection of the spike burst activity for even the weakest contractions (≈10–20 dB) (Staude et al 2001).

The average difference in SNR when comparing spike bursts measured with the Intsy system configured with 1 kHz versus 100 Hz sampling rates was 5.6  ±  0.2 dB, or equivalently a 1.9  ×  difference in signal strength. The smaller SNR with 100 Hz sampling may be primarily attributed to the fact that roughly half of the power lies above the 50 Hz Nyquist frequency in the canonical sEMG power spectrum, so the slower sampling rate can not properly capture the maximum deflections of underlying high frequency activity.

In all instances the Intsy system recorded high quality sEMG signals in an every day office environment. When configured for the higher (1 kHz) sampling rate, the new Intsy system signal quality was comparable to that of Biosemi, and was also high enough at a lower sampling rate (100 Hz) to still enable accurate data analyses.

We have developed and validated a new bioampilfier module, based on the Intan RHD2132 electrophysiology measurement chip and the Teensy 3.2 microcontroller, which is capable of making high-quality recordings with comparable signal quality to commercial systems that cost up to  ≈$10\times$ more. Intsy's combination of features—maximum channel count, real-time wireless data streaming capability, high bandwidth over full-speed USB, and low cost—offers marked advantages over other modules in its class, as summarized in table 1. To the best of our knowledge, this is the first validated, open-source system with real-time wireless streaming capability across  ⩾32 channels. Thus, Intsy now offers a suitable and attractive option for use in various bioelectric measurement domains. In addition, as the Intsy sytem hardware and software is open-source, it may be customized by the end-user for a particular application.

While the Intan RHD2132 chip has been recently incorporated into custom neural optogenetics devices (e.g. Nguyen et al (2014) and Gagnon-Turcotte et al (2017)), this paper now presents a novel application and validation of the Intan chip for measurements of GI electrical activity, a field of rapidly-growing clinical interest with a critical need for a reliable, affordable recording system (O'Grady et al 2014).

As the Intsy system was shown to record serosal gastric SWs with high fidelity, proposed further uses in the GI electrophysiology domain include interfacing with a 32–64 channel endoscopic recording device (Angeli et al 2017), measuring the cutaneous HR-EGG in ambulatory patients (Gharibans et al 2017), and tracking the progression of post-operative ileus using an array of cutaneous electrodes (Vather et al 2014).

Given the positive results from the sEMG validation study in this paper, another potential application in the GI domain is anorectal electromyography, which has been used previously to assess muscle activation and relaxation and provide biofeedback therapy for patients suffering from fecal incontinence (Bharucha and Rao 2014). Other potential applications outside of the GI domain include studies involving the 12-lead ECG, 32-channel EEG, multielectrode mapping of propagating electrical activity in the urinary bladder (Hammad et al 2014), or the 64-channel electrohysterogram (EHG) (Rabotti et al 2010, Rabotti and Mischi 2015), all of which are areas of current research with substantial potential for clinical application. The Intsy system is also generally suitable for student laboratory electrophysiology experiments.

Before adopting the Intsy system for a study, possible limitations should be carefully considered. For instance, gastric SW activation front mapping relies on detection of the large downstroke component only, but is insensitive to the precise waveform shape otherwise. Hence, SW activation front mapping results obtained with the Intsy and Biosemi systems were essentially identical. Given the typical readily identifiable triphasic gastric serosal SW waveform recorded by Intsy, it might also be utilized for the recently introduced GI activation recovery mapping technique (Paskaranandavadivel et al 2017b)

The total baseline noise level of the Intsy system (≈6–8 μV RMS) compares reasonably well to a typical commercial system, but is slightly higher by a few μV (Huigen et al 2002). This renders Intsy less suitable for applications where the signal strength is expected to be very low ($\lesssim$50 μV). With no passive or active shielding incorporated, Intsy recordings may be susceptible to contamination by power line interference, the negative effects of which can be mitigated by carefully choosing an experiment location away from interference sources, implementing passive shielding, maintaining short cable leads, and applying digital filter techniques (Clancy et al 2002).

In our experience, the maximum stable wireless data transmission bandwidth was 110 Hz  ×  32 channels (70% of the theoretical maximum bandwidth for a 115200 baud rate). This sample rate was sufficient for identifying the low-frequency gastric SW activity with high fidelity. It was also sufficient for some higher frequency applications, such as identifying higher-frequency sEMG on/off times as well as identifying trends in signal strength versus force of contraction. However, the wireless data bandwidth is not high enough to accurately determine the full sEMG power spectrum or make high-fidelity recordings of high-frequency components, which may be critical for some research applications. Therefore, the application scope of Intsy in the sEMG domain is limited, in general, to clinical applications for which relatively low sampling rates are sufficient.

In the current design, we opted for Bluetooth over WiFi based on the tradeoff between maximum data streaming rates and power consumption. A future version of the Intsy hardware module could also incorporate an off-the-shelf WiFi module, permitting much higher data streaming rates, with a concomitant  ≈$3\times$ increase in power consumption. Additionally, a future version could enable higher per-channel data rates across fewer channels with Bluetooth by selecting a subset of channels for A/D conversion and data transmission; 1600 S/s/ch could be achieved for two channels, for example.

The maximum USB wired data transmission rate of 2 kHz  ×  32 channels may not be high enough for projects in the neural field where measuring and clustering individual spikes is critically important. This issue may be at least partially alleviated by using a microcontroller with a faster CPU capable of implementing the maximum Intan RHD2132 SPI port communication rate (24 MHz), thereby reducing the ADC scan time by half. A future version of the Intsy device will integrate a microSD card port that can be used to store data in the event that the wireless data link breaks (Springston et al 2016). SD card storage would also allow increased portability for ambulatory studies; the test subject would not need to remain within  ≈10 m of the host computer.

Additional future revisions will decrease power consumption, physical size and cost of the Intsy hardware module. For Intsy's current design, we opted to use through-hole components for ease of assembly, intending that anyone with even modest experience with electronics assembly could make a module. Using surface mount components would allow for the module to have smaller dimensions. Doing so would also allow the RHD2132 chip(s) to be integrated directly onto a custom PCB, obviating the need for SOIC-to-through-hole adapters, while also reducing the total cost by nearly one-half.