3.1 Chamber’s Spatial Characteristics

To simulate MEMS microphone installation spaces, we used 3D modeling to design experimental chambers with equal volume and different sizes of holes. Fig. 2 shows a 3D model of the experimental chambers, and Fig. 3 shows samples produced using a 3D printer. The chambers were designed to be thin and small to reflect the dimensions of smartphones and laptops, where MEMS microphones are commonly installed. The chambers measured 20 ${\times}$ 20 ${\times}$ 3.5 mm$^{3}$ when a bottom panel mounted with a MEMS microphone was adjoined. The chambers were punched with holes of 1, 1.5, and 2-mm diameters.

Fig. 4 shows a cross-sectional diagram of the chambers with a MEMS microphone installed, which reflects the structure of chambers used for practical applications. The structure is akin to a Helmholtz resonator with a narrow hole and even internal volume. Therefore, the effectiveness of the design can be verified by comparing the experimental results to the maximum resonant frequency calculated using the equations for a Helmholtz resonator ^[3].

For the calculation, we established an equation for the internal volume, V = V1–V2, where A is the cross-sectional area of the input hole in Fig. 4, and 1 is the length of the hole. The pressure change outside the hole $\textit{Po}$$^{(t)}$ is caused by the air mass Q (kg/s) drawn into volume V per second. In addition, $\textit{Pi}$$^{(t)}$ is the pressure change inside volume V. When the hole resonates, the pressure change in the input part is very subtle, while the air inflow and outflow are maximized. Therefore, the frequency that minimizes the pressure change at the hole entrance $\textit{Po}$$^{(t)}$ is the maximum resonance point. The frequency that minimizes the pressure in the hole entrance is given by:

$\textit{c}$ is the velocity of sound in the measured space. To reflect the experimental environment, c was modified using:

and

The velocity of sound in the air is expressed in Eq. (2), where ${\gamma}$ is the adiabatic index, $\textit{R}$ is the gas constant, $\textit{T}$ is the temperature, and $\textit{M}$ is the air mass. It can be assumed that all of these terms except for the temperature are constant under the experimental conditions. Hence, entering the temperature in degrees Celsius gives the velocity of sound at the target temperature. The temperature was set to 22$^{\circ}$C, as shown in Eq. (3). The resonant frequencies for each diameter were obtained by substituting the calculated sound velocity into Eq. (1) along with the other values shown in Table 1. The results were rounded to one decimal place.

Fig. 2. 3D models of the experimental chambers.

Fig. 3. Photograph of the experimental chambers produced using a 3D printer.

Fig. 4. Cross section of a chamber with a MEMS microphone installed.

3.2 Frequency Response Experiment

A MEMS microphone was installed in each of the experimental chambers, and the frequency response was recorded. The measurements were conducted in a complete anechoic chamber under ISO 3745 standards. An air-handling unit was used to maintain the target temperature of 22$^{\circ}$C. The microphone measurement method was used with a standard reference microphone for IEC 60268-4 ^[4,5].

Fig. 5 shows a block diagram of the equipment and the experimental facility. The equipment included an Audio Precision APx555 signal analyzer, B&K 4191 reference microphone, Focal CHORUS-V 806V reference speaker, B&K 2690 conditioning amplifier, Crown XLi 2500 power amplifier, and Agilent E3632A DC power supply. The MEMS microphone used for this experiment was a bottom-port-type capacitive analog microphone unit, as shown in Fig. 6. The drive voltage was 3 V. The microphone was installed in the chambers as shown in Fig. 4. It was connected to the input/output and voltage load lines and covered with an airtight seal using rubber glue to prevent gaps.

The frequency responses for the chambers with holes of different diameters are illustrated in Fig. 7. The solid, thick dotted, and thin dotted lines represent the chambers with holes that were 1, 1.5, and 2 mm in diameter, respectively. The resonant points of each line have bandwidths comparable to those calculated in Table 1.

The effectiveness of the sample sets can be verified using the difference between the calculated and measured frequencies. The difference will be greater if the chambers are improperly designed or the MEMS microphone is malfunctioning. We produced a filter to compensate for the change in frequency in the opposite phase based on the measured frequency response characteristics. The compensation bandwidth was set to the primary bandwidth of the voice field, which ranges from 100 Hz to 4 kHz with a maximum of 20 dB considering the actual limit of the DSP gain.

A digital filter was applied to the signal. The frequency response characteristics were adjusted through a digital equalizer for the frequency modulated by the spatial characteristics. Based on the measurement result, the core of the implementation is to make detailed factors of digital equalization suitable for the voice bandwidth and apply them to the input gain of the DSP.

Fig. 5. Equipment and experimental facility.

Fig. 6. Bottom port-type capacitive analog microphone unit used as the MEMS microphone.

Fig. 7. Frequency responses for chambers with holes of different diameters.

3.3 Voice Quality Improvement Experiment and Result

To verify the improvement in voice quality that was obtained using the compensation filter, we applied a digital filter to the input/output path of the experimental signal. The frequency response filters obtained by simulating the chamber’s characteristics were applied to the input terminals. The compensation filters were designed to adjust for the primary bandwidth in the opposite phase and applied to the output terminals.

The digital filter is responsible for restoring the frequency modulation occurring in the main band that affects the intelligibility of speech. The recovery of frequency modulation through this signal processing has two advantages. First, unlike conventional analog circuit systems, there is no effect of electrical signal cancellation due to a phase difference. Furthermore, since there is no additional hardware, it also reduces cost.

Fig. 8 shows the digital filters that compensate for the frequency responses of the primary bandwidth. They were produced with consideration of the DSP’s application range and bandwidth. The solid, thick dotted, and thin dotted lines represent the compensation filters for holes that were 1, 1.5, and 2 mm in diameter, respectively. The improvement was measured using perceptual objective listening quality assessment (POLQA) with and without the compensation filters.

The speech transmission index (STI) evaluates voice quality using noise generation in an even narrowband. In contrast, POLQA is a voice quality testing method that uses subjective perception and evaluates quality by reflecting perceptual characteristics based on a voice-sound source. We used the POLQA V2.4 algorithm based on the ITU-T P.863 (2014) standards. The results are shown in Fig. 9 using MOS-LQO, which denotes the mean opinion score of listening-only quality and objective. The score ranges from 1 to 5, with higher numbers representing higher quality ^[6].

The POLQA test results for each hole diameter are shown in Table 2. For each diameter, the score increases when the filter was applied. Given the sensitive reaction to the diameter of the holes, the performance of the compensation filter can be improved though testing and estimation of spatial characteristics.

This difference is due to the superposition of digital filters that can be applied to the DSP without the need for additional sensors or components. Although it depends on the performance of the DSP processor, considering that multiple digital filters can be applied by overlapping, it can cope with frequency modulation from a more complex type of space. Existing intelligibility optimization work did not understand the frequency modulation of the structural space. Because of this, many trials and errors occur because the correct standards are not met, and this test result shows a way to shorten the time to worry about.

This study allows us to grasp the structural characteristics of the space where MEMS microphones are installed, and provides an improvement method for this in the area of digital signal processing. It analyzes the structural space and optimizes the physically unmodifiable parts using digital processing. This can produce high effects with low cost. Therefore, it has sufficient industrial value for manufacturers who manufacture or use MEMS microphone components. In addition, researchers studying MEMS microphones are expected to have sufficient value as an application technology for technology development because they can remove the effect of frequency modulation due to spatial effects and study the characteristics of the sensor.

Fig. 8. Digital filters that compensate for the frequency responses in the important bandwidth.

Fig. 9. MOS-LQO scores with a filter (top) and without one (bottom) for a chamber with a 1-mm-diameter hole.