GT1M and GT3X

The AM7164 was replaced by the GT1M in the mid 2000s. ActiGraph stated that the GT1M was more cost-effective and had upgraded technology while providing similar output to its predecessor. The main difference between the AM7164 and the GT1M is that the former has a piezoelectric sensor that detects dynamic accelerations (resulting from motion), while the GT1M uses a capacitive accelerometer capable of detecting both static (e.g., force of gravity detected when stationary) and dynamic accelerations. Unlike piezoelectric accelerometers where a change in acceleration results in the production of a proportional electric charge, a capacitive accelerometer detects change in acceleration through changes in capacitance of the sensing element; in other words, variations in the sensor’s electric charge storage potential.

The acceleration sensor in the GT1M is the ADXL320 (Analog Devices, Norwood, MA), which is a 4 × 4 × 1.45 mm surface micromachined, monolithic integrated circuit chip (polysilicon), dual-axis Microelectro-Mechanical-System (MEMS) accelerometer. The ADXL320 has a full-scale range of ± 5g and measures both static and dynamic accelerations. However, ActiGraph-specified restrictions allow the detection of only those accelerations that lie between 0.05 to 2 g. Structurally, this capacitive accelerometer has two independent polysilicon fixed plates that act as electrodes, and between the two fixed plates is a parallel movable central plate. This plate is movable because it is suspended as a bridge structure using polysilicon springs that are attached to the sides of the structure. The suspended central plate forms two back-to-back capacitors with each fixed plate on either side. The three plates form a differential capacitor in which the deflection of the central plate in response to acceleration causes a change in the capacitance of the two back-to-back capacitors. The change in capacitance causes specific voltage changes to the existing electric flow, which is an analog signal that is proportional to the detected acceleration. This low voltage signal is first amplified and is then digitized, rectified, and the direction/s (vertical, antero-posterior, and medio-lateral axes) of the acceleration are determined using 12-bit ADC (sampling rate= 30 Hz.) and phase demodulation techniques. Finally, the signal is filtered at a bandwidth of 0.25 to 2.5 Hz. This means that the acceleration signal is markedly attenuated if the frequency of acceleration peaks falls outside this range. ActiGraph’s rationale for using a band-pass filter is that most forms of human movement fall within this frequency range, and the filter might be useful for eliminating high-frequency artifact vibrations.

In mid-2009, ActiGraph released the GT3X containing an ADXL335 accelerometer (Analog Devices, Norwood MA). The ADXL335 is a 4 × 4 × 1.45 mm triaxial capacitive MEMS sensor with a full scale range of ± 3g. The structure and theory of operation of the ADXL335 is similar to that of the ADXL320. Other than the accelerometer in the GT3X, all other specifications, including the filter and the ADC, are precisely the same as in the GT1M. Additionally, although both the GT1M and the GT3X have the capability to measure static accelerations, only the GT3X provides inclinometer output. The GT3X uses vector magnitude data from all three axes and assigns a number from 0 to 3 to distinguish whether an individual is not wearing the monitor (number 0), is standing (number 1), lying (number 2), and sitting (number 3).

ActiGraph Pre-filtered Raw-Mode Output

Unlike the AM7164, both the GT1M and the GT3X also provide output in the pre-filtered raw- mode. This is an important feature in the two capacitive activity monitors because the general consensus at the July 2009 “Objective Measurement of Physical Activity: Best Practices and Future Directions” conference was that researchers should begin using raw accelerations instead of activity counts when measuring physical activity. Before April 2010, the output in the pre-filtered raw mode was a representation of the analog-to-digital quantization of accelerations sampled every 0.033 s (sampling frequency of 30 Hz.) by the accelerometer. The analog-to-digital quantization values are discrete, unsigned digital numbers proportional to the voltage signal generated by the accelerometer in response to acceleration. Firmware modifications in April 2010 changed the output in the pre-filtered raw mode to represent the actual g-force sampled every 0.033 s.

To obtain these “raw-mode” variables, the analog-to-digital quantization is calculated using several parameters, including the sensitivity of the ADXL320/335 accelerometers (174 millivolts/g for each axis), the total number of levels in which a sampled acceleration can be classified (12-bit analog to digital converter enables 4096 (2¹²) different levels), and the zero g offset of 1.5 V. The 4096 levels of the 12 bit ADC is spread equally across a 3 V reference voltage. Therefore, each ADC step is equivalent to 732.4 μV. Taking into account the zero-g offset of 1.5 V, the voltage signal when the accelerometer detects an acceleration of + 3 g would be 2.022 V ([3 g * 0.174 V/g] + 1.5 V). Therefore, the pre-filtered raw-mode output equals 2760 [(2.022 × 1000 × 1000) μV÷ 732.4 μV]. The raw-mode output in g force units (e.g., 3 g) is obtained in the following manner: 2.022 V (voltage signal from accelerometer) −1.5 V (zero-g offset) ÷ 174 millvolts/g (sensitivity of accelerometer)

Although the ADXL320 and the ADXL335 originally have unequal sensitivities of 174 millivolts/g and 270 millivolts/g, respectively, ActiGraph states that the sensitivity of the GT3X accelerometer is scaled to match that of the GT1M to ensure comparability in output. However, it is unclear if a GT1M initialized in the pre-filtered raw-mode will measure a higher range of accelerations (−5 to +5g) than the GT3X, which only measures accelerations in the range of −3 to +3 g. If this is the case, the output from the GT1M and GT3X in pre-filtered raw mode will not be similar to each other. Output in the post-filtered raw-mode is not a representation of the actual raw acceleration signal because it is subjected to transformation associated with manufacturer specifications (for example-band-pass filtering).

Firmware Updates

Firmware is defined as a software programming code containing a set of instructions that enable communication between the hardware components of a device. The firmware in the GT1M and the GT3X is stored on the onboard flash read-only-memory and it performs several functions, including filtering the accelerometer signals. In comparison, this function in the AM7164 was performed by hardware components. Because the firmware is stored within flash memory in the GT1M and the GT3X, it can be erased and replaced with updated versions. Scheduled firmware releases by ActiGraph usually coincide with improvements that are made to the devices. For example, firmware version 03.01.01 for the GT3X was a scheduled release containing added support for the new 16 Mb memory devices. Similarly, firmware release 04.01.00 in July 2008 for the GT1M enabled users to measure accelerations in the antero-posterior axis in addition to the vertical axis. On the other hand, unscheduled firmware releases are also made when bugs” (e.g., an issue that causes inaccuracy of output) are identified and need to be fixed. A potential bug we observed in our lab was the presence of significant differences among activity counts in the antero-posterior axis from the GT3X and the GT1M (firmware 04.01.00) at different walking and running speeds (unpublished observation). This problem was particularly evident at the speed that marked the transition from walking to running (unpublished observation). A subsequent firmware release for the GT1M resolved this likely bug by restoring the similarity in antero-posterior axis activity counts during walking and running between the two monitors.

Although susceptible to programming errors, an advantage of having a firmware-based system is the reduction in possible errors due to the loss of calibration. As compared to the firmware filter in the GT1M and the GT3X, the drift and tolerance characteristics of each of the multiple semiconductor components in the AM7164 hardware filter contribute to the possibility of a loss in calibration. ActiGraph states that the GT1M and the GT3X do not require calibration. However, in our lab we have observed that the GT1M and the GT3X can lose calibration (unpublished observation). Researchers are encouraged to use the option available in the ActiLife software that allows one to confirm calibration of the GT1M and the GT3X.

Because the GT1M and the GT3X are used to measure physical activity, a possible way for these monitors to lose calibration is from exposure to impact forces, such as being dropped. Several capacitive accelerometer manufacturers strongly caution against dropping the accelerometer onto hard surfaces and also furnish drop-test details in the technical specifications. Even though the electronics of the GT1M and GT3X are enclosed in a hard polycarbonate plastic casing, we were curious to determine whether dropping a calibrated GT1M on a hard surface would cause the ADXL accelerometer within to lose calibration. To answer this question, we dropped a calibrated GT1M a few times on a concrete floor from a height of approximately 2.0 m. We selected this height because we were not aware of the ADXL320 accelerometer’s absolute maximum rating for a drop test, and the height of 1.8 m was the absolute maximum rating height for a similar capacitive accelerometer that we use in our lab. Dropping a GT1M from this height did not cause the GT1M to lose calibration. However, we cannot say whether calibration would be compromised if the monitor were repeatedly exposed to similar forces.

We thank Dr. David R. Bassett for his contributions to this paper.