sensor for urban noise monitoring
Academic year 2019-2020
Master of Science in Electrical Engineering - main subject Electronic Circuits and Systems Master's dissertation submitted in order to obtain the academic degree of
Counsellor: Ir. Vincent Spruytte (ASAsense)
Supervisors: Prof. dr. ir. Bert De Coensel, Prof. dr. ir. Dick Botteldooren
Student number: 01404010
sensor for urban noise monitoring
Academic year 2019-2020
Master of Science in Electrical Engineering - main subject Electronic Circuits and Systems Master's dissertation submitted in order to obtain the academic degree of
Counsellor: Ir. Vincent Spruytte (ASAsense)
Supervisors: Prof. dr. ir. Bert De Coensel, Prof. dr. ir. Dick Botteldooren
Student number: 01404010
First and foremost, I would like to express my gratitude to prof. dr. ir. Bert De Coensel and prof. dr. ir. Dick Botteldooren for giving me the chance to fulfil this master thesis at the Waves group. I want to thank them for sharing their expertise and giving advice through this year. In addition, I would like to thank dr. ir. Joris Van Kerrebrouck for giving me instructions and useful tricks for soldering the microscopic parts of the PCB.
Furthermore, I would like to give special thanks to Bram Vandekerckhove, Nicolas Claus and Fien Vanden Hautte for their friendship and support through my entire studies. Finally, I want to show gratitude to my family and especially my brother for offering a listening ear and encouraging me.
The author gives permission to make this master’s dissertation available for consultation and to copy parts of this master’s dissertation for personal use. In the case of any other use, the limitations of the copyright have to be respected, in particular with regard to the obligation to state expressly the source when quoting results from this master dissertation.
De auteur geeft de toelating deze masterproef voor consultatie beschikbaar te stellen en delen van de masterproef te kopi¨eren voor persoonlijk gebruik. Elk ander gebruik valt onder de beperkingen van het auteursrecht, in het bijzonder met betrekking tot de verplichting de bron uitdrukkelijk te vermelden bij het aanhalen van resultaten uit deze masterproef.
by
Margaux Van Kerckhove
Master’s Dissertation submitted to obtain the academic degree of Master of Science in Electrical Engineering
Academic 2019–2020
Promoters: Prof. dr. ir. Bert DE COENSEL, Prof. dr. ir. Dick BOTTELDOOREN Counsellor: ir. Vincent Spruytte (ASAsense)
Faculty of Engineering and Architecture Ghent University
Summary
An analog sound level sensor is designed and implemented on a PCB. It consists of a microphone FG-23329-P07, amplification circuit, an A-weighting circuit and an integrating circuit, that executes fast and slow time weighting. The dynamic range of the sensor reaches from 35 to 105 dB and meets the overall accuracy of a type 2 sound level meter, except for signals around 70 dB where the error is slightly bigger than allowed. For frequencies above 200 Hz the A-weighting, time weighting and ripple specifications meet the requirements of a type 1 sound level meter. The design has a power consumption of 206.5 mW and a cost of 204.85 euros. The sound level sensor can be used to measure urban noise in smart city or citizen science applications.
Keywords
sensor for urban noise monitoring
Margaux Van Kerckhove
Supervisors: prof. dr. ir. Bert De Coensel, prof. dr. ir. Dick Botteldooren Abstract—An analog sound level sensor is designed and implemented on
a PCB. It consists of a microphone FG-23329-P07, amplification circuit, an A-weighting circuit and an integrating circuit, that executes fast and slow time weighting. The dynamic range of the sensor reaches from 35 to 105 dB and meets the overall accuracy of a type 2 sound level meter, except for signals around 70 dB where the error is slightly bigger than allowed. For frequencies above 200 Hz the A-weighting, time weighting and ripple specifications meet the requirements of a type 1 sound level meter. The design has a power consumption of 206.5 mW and a cost of 204.85 euros. The sound level sensor can be used to measure urban noise in smart city or citizen science applications.
Keywords—analog design; acoustic sensing; smart cities; time weighting; noise monitoring; low-power;
I. INTRODUCTION
F
REQUENT exposure to unwanted noise can cause audi-tory and audiaudi-tory damage to human health. The non-auditory damage, mostly caused by environmental noise, can lead to cardiovascular disease, stress, sleep disturbance and a decrease in cognitive performance [1]. This problem mainly oc-curs in cities. To reduce the effects of the urban noise on the public health, a sound analysis of the city has to be made. An efficient way to make a sound map of the city is by applying the concept of a Smart City, where several sound level sensors are spread over the city. These sensors are connected to a wireless network that gathers the measurements in time and space. An-other way to construct noise maps of cities, is to involve the cit-izens in citizen science projects. In this way, large-scale maps can be formed by the citizens that gather data by using their smartphone as a sound level sensor. If mobile crowd sensing is applied, the cost of these projects is rather low because there is no need for extra technical infrastructure. However, the mea-surement results are not always that accurate. Therefore, re-search on low-cost sound level sensors with a high accuracy is necessary.Sound levels are more difficult to measure than other quantities, such as temperature or air pollution, because a sound pressure signal varies fast over time and contains a lot of information. The sound pressure level is calculated by using formula 1 and is expressed in dB SPL [2].
Lp= 20∗ log10( p
pref) with pref= 2∗ 10−5 N m2 (1) To obtain useful measurements, a sound level meter has to av-erage the signal over time in energy equivalent levels. The hu-man ear is only sensitive to frequencies between 20 and 20 kHz, therefore a sound level meter must also filter the frequencies and adjust the level in such a way that the measurements are compa-rable with how the human ear would receive the signal. Over the years, the sound level meter evolved from a heavy and
inaccurate fixed analog model to a hand held measurement in-strument. With the arrival of the digital sound level meters it was possible to integrate measurements over a longer period of time and display the results in real time. However, the disad-vantage of the digital sound level meter is that it needs a lot of power to obtain accurate calculations in the lower frequencies. If this sound level meter is used in large scale applications it is not practical to change its batteries frequently.
For various cities, research has already been conducted into the optimal design of a sensor for urban noise monitoring. In the SONYC [3] project they used a Raspberry Pi single-board com-puter in combination with a MEMS microphone and a Wi-Fi antenna. This is a low-cost design with a total cost per sensor of 73.3 euros, but it is has a power consumption of about 2.5 Watt. Another recent study of Fernandez-Prieto et al. [4] describes a system where nine sensor nodes are used. Each node consists of an Arduino Due with antenna, an electret microphone and a mi-crophone amplifier. These sensors have a power dissipation of 421 mW per sensor without taking into account the power con-sumption of the Arduino device. The cost of a device is about 110 euros.
To compare the measurement results of different types of sound level meters, international standards are defined. The two most used standards for sound level meters are the ANSI (American National Standards Institute) and IEC standard (International Electrotechnical Commission). To check the specifications of the design described in this paper, the ANSI S1.4 standard [2] is used. This standard is very similar to the IEC 61672 standard [5] and defines the requirements for the performances of sound level meters. Depending on the accuracy, the meters are classi-fied in type 0, type 1 or type 2 for the ANSI standards and in class 1 or class 2 for the IEC standards.
In this paper the design of an analog, low-power and low-cost sound level sensor is discussed. In the next section the different parts of the design are specified. The results of the simulations and the measurements are given in section 3. This paper ends with a conclusion, in which the improvements on the design and possibilities for further research are discussed.
II. DESIGN
The sound level sensor consists of 5 parts: the microphone, an amplification circuit, an A-weighting circuit, an integration circuit and a processing part.
A. Microphone FG-23329-P07
The microphone for this design is chosen based on earlier re-search of the waves group [6]. The FG-23329-P07 is an omni-directional electret condenser microphone that can operate in a
frequency range of 100 Hz to 10 kHz. The sensitivity is -53 dB at 1 kHz with reference to 74 dB SPL (-33 dB with reference to 94 dB SPL) and the noise floor of this microphone is 30 dB SPL at 1 kHz. The assumed dynamic range of the microphone reaches from 30 to 110 dB, which can be converted to voltage levels of 13.9 µV to 139 mV. The microphone has a small inte-grated preamplifier which needs an external power supply of 1.3 V.
B. Amplification circuit
The electrical output signal of the microphone has to be am-plified to an operational level for the integrating circuit. To op-timally maintain the dynamic range, the signal levels are split up in low signal levels and high signal levels, as can be seen in figure 1. The low signal levels are amplified with a factor 500 and the high signal levels with a factor 5. The amplification is done by using multiple stages of non-inverting opamps. To minimize the extra noise added to the signal due to this ampli-fication circuit, an opamp with a low off-set voltage is chosen. The OPA735 is optimized for low signal levels and has an off-set voltage of maximally 5 µV. This opamp has a quiescent current of only 750 µA and needs a 5 V supply which results in a rela-tively low power consumption of 3.75 mW.
C. A-weighting
The A-weighting curve is approximately the inverse of the 40 Phon equal loudness contour [7]. In the ANSI S1.4 standard [2] the A-weighting transfer function is defined by a number of poles. Multiple options for the A-weighting circuit have been explored but since this should be a low power design, complex multiple Chebyshev filters are not suitable. The A-weighting circuit for this design is based on the circuit of the Sennheiser audio level meter UPM 550 [8]. It consist of multiple stages of RC filtering and one non-inverting opamp to adjust the gain so that the transfer function passes 0 dB at 1 kHz. The calculated transfer function of the circuit is given in figure 2.
The 6 complex poles of this transfer function are compared to the poles of the standard A-weighting in table I. As can be seen, they more or less correspond except for the fact that the double poles of the standard A-weighting are split up in two single poles.
D. Integration
The integration circuit converts the output of the frequency weighting circuit into a DC value. This is done by taking the ab-solute value of the input signal, squaring it and applying a time weighting. The integrated chip that is chosen for this part is the AD636. This chip is optimized for low signal levels between 0
Fig. 2: Transfer function of the A-weighting circuit. Transfer function UPM 550 ANSI standard
Poles [Hz] Poles [Hz] p1 7.32 20.6 p2 40.6 20.6 p3 97.6 107.6 p4 756.6 737.9 p5 9687 12194 p6 15551 12194
TABLE I: Comparison of the poles of the UPM 550 circuit and ANSI standard.
and 200 mV rms and it has the option to read out a dB value directly. The dynamic range of the chip is 50 dB. It is possible to supply the AD636 with a battery of 9 V and the typical quies-cent current is only 800 µA, which results in a very low power consumption of 7.2 mW. This single supply operation can be realized by placing the ground in the middle of the circuit and using a bandgap voltage reference chip (AD589) that provides a fixed 1.2 V reference level. The time weighting is controlled by the value of a capacitor: a time constant of 25 ms equals 1µF. In this design, both fast time weighting (125 ms) and slow time weighting (1000 ms) are implemented.
E. Processing
In the processing part the output value of the integrator of either the low signal part or the high signal part is selected. As the complexity, cost and power should be low, an Arduino Nano is proposed as processing board.
III. RESULTS
After various simulations in LTspice, the design was fabri-cated on a FR4 PCB. The components were manually soldered on the PCB. To measure the PCB, an external sound card con-nected to a PC, a voltage source and an oscilloscope are used. A. Amplification
The amplification was simulated and the desired amplifica-tion factor was obtained for both the low signal part and the high signal part. The simulation result of the amplification of the low signal levels is given in figure 3, the input signal is amplified with a factor 500 and the output signal is not distorted. As there was a mistake in a connection of the amplification circuit, it was
impossible to have good measurements of the amplification cir-cuit on the PCB.
B. A-weighting
To simulate the A-weighting circuit, a linear sweep from 1 Hz to 20 kHz was applied at the input of the circuit. The error be-tween the result of this simulation and the standard A-weighting (according to ANSI S1.4 [2]) was smaller than the allowable er-ror of a type 0, type 1 and type 2 sound level meter. To measure the A-weighting circuit on the PCB, an exponential sweep was used as input signal. The transfer function could be calculated
Fig. 4: Input and output signal of the A-weighting circuit with exponential sweep measurement.
from the recorded input and output signal (see figure 4) using the method described in ([9]). The result is given in figure 5. As can be seen, the transfer function of the simulation approaches the standard A-weighting very well. The transfer function shows some large deviations in the low frequencies. This may be due to a lack of energy in the low frequencies of the exponential sweep. The output of the circuit for frequencies above 200 Hz has deviations that are within the limits of a type 1 sound level meter.
C. Integration
The dB output of the AD636 is expressed in mV: 1 dB cor-responds to 3 mV and as the amplitude of the input signal increases, the integrator output becomes more negative. The AD636 is calibrated at a level of 77 mV rms. When a sinusoidal signal with a frequency of 1 kHz and amplitude 77 mV is ap-plied as input signal of the integrating circuit, the output level is
-0.37 dB with an error of -1.2dB/+0.42 dB. This error is slightly too much to meet the requirements of a type 2 sound level meter.
Time weighting
To test the time weighting, burst signals with different duration were used as input signal. The burst signal is generated by mul-tiplying a sinusoidal signal with a Tukey window, which gradu-ally suppresses a small part of the beginning and ending of the sinusoidal signal (see figure 6). The deviation of the output of
Fig. 6: Burst signal of 200 ms.
the integrating circuit between using a burst signal or a continu-ous sinusoidal signal as input is then calculated. The results are shown in table II, from which it can be deduced that the circuit meets the requirements of a type 1 sound level meter. The re-sponse of the integrator on a 200 ms burst is given in figure 7 for fast time weighting. Out of these simulation tests, it is also verified that the output should decreases with 10 dB in at least 0.5 s for fast and in 3 s for slow weighted signals.
Ripple
The ripple on the output signal determines the accuracy of the simulation results. In table III the ripple for sinusoidal input signals with frequencies of 100, 1k and 10k Hz is given. This ripple is larger at lower frequencies than at higher frequencies. The ripple can be reduced by increasing the time weighting con-stant or by inserting a post-pole filter. As can be conducted from the simulation results, signals with a frequency lower than 200 Hz will give output values that do not meet the requirements of the ANSI standard.
Duration Difference with Allowable
Weighting burst continuous difference
(ms) signal (dB) standards FAST 200 1.3 1 50 1.33 4.8 20 1.37 8.3 5 7.2 14.3 SLOW 2000 1.2 0.6 500 1.45 4.1 200 1.4 7.4 50 1.3 13.1
TABLE II: Comparison of the deviations in output between burst signals of different duration and a continuous signal.
Dynamic range
The dynamic range of the integration circuit was tested by ap-plying sinusoidal signals of 1 kHz with different amplitudes at the input. According to the datasheet of the AD636 the optimal range should be 0 to 200 mV rms. In LTspice, it was possible to have accurate results for input levels up to 2.2 V rms. When the same test was executed on the PCB, the AD636 only gives right output values for input levels up to 400 mV rms. The output of the PCB measurements is compared in figure 8 to the results of the spice simulations and the expected output.
An overview of the complete signal flow is given in table IV The S1.4 ANSI standard specifies that the overall error for a sinusoidal signal can be ±1.6 dB for a type 1 sound level meter and ±2.3 dB for a type 2 sound level meter. This design meets the type 2 requirements for signal levels between 35 dB to 65
Frequency [Hz] FastRipple on output signalSlow 100 Hz 33.7 mV (11.2 dB) 28.68 mV (9.56 dB)
1 kHz 5.36 mV (1.79 dB) 4.45 mV (1.48 dB) 10 kHz 1.28 mV (0.43 dB) 1.02 mV (0.34 dB) TABLE III: Ripple on output signal integrator for different fre-quencies and different time weightings.
LOW SIGNAL LEVELS
Microphone Amplifi- Output Expected Error Error Output Output cation integrator output (-) (+)
(dB) (mV) x500 (mV) (dB) (dB) (dB) (dB) 30 0.0139 6.96 17.2 20.4 -3.6 35 0.0247 12.37 14.1 15.4 -1.88 40 0.044 22 9.87 10.4 -1.28 +0.18 45 0.0782 39.1 5.2 5.4 -1.02 +0.58 50 0.139 69.6 0.39 0.4 -0.85 + 0.79 55 0.247 124 -4.48 -4.6 -0.73 +0.95 60 0.44 220 -9.32 -9.6 -0.58 +1.12 65 0.782 391 -14.2 -14.6 -0.45 +1.32
HIGH SIGNAL LEVELS
Microphone Amplifi- Output Expected Error Error Output Output cation integrator output (-) (+)
(dB) (mV) x5 (mV) (dB) (dB) (dB) (dB) 70 1.39 6.96 17.2 20.4 -3.6 75 2.47 12.37 14.1 15.4 -1.88 80 4.4 22 9.87 10.4 -1.28 +0.18 85 7.82 39.12 5.2 5.4 -1.02 +0.58 90 13.9 69.6 0.39 0.4 -0.85 +0.79 95 24.7 124 -4.48 -4.6 -0.73 +0.95 100 44 220 -9.32 -9.6 -0.58 +1.12 105 78 391 -14.2 -14.6 -0.45 +1.32
TABLE IV: Overview of signal flow.
dB and from 75 dB to 105 dB. D. Power and cost
Based on the LTspice simulations and the data sheets of the components, an estimation of the power consumption of this de-sign could be made. The total power consumption of the dede-sign is 206.5 mW. The Arduino nano consumes 95 mW and is thus responsible for almost half of the total power consumption. The cost of the design is 204.85 euros if one PCB was fabri-cated and in case 100 or 1000 PCB’s are fabrifabri-cated, the cost is respectively 161.8 or 134.8 euro per PCB. The cost is higher than the projects ([3] and [4]) discussed in the introduction, but the power consumption of this design is considerably lower.
IV. CONCLUSION
For frequencies above 200 Hz, the A-weighting, time weight-ing and ripple specifications meet the requirements of a type 1 sound level meter. The dynamic range of the sound level meter reaches from 35 to 105dB. The error on the output of the sound level sensor meets the standards for a type 2 sound level meter
dB to 100 dB, which in turn results in a total range of 35 dB to 100 dB. It is a trade-off that has to be made between the width of the dynamic range, the accuracy and the cost. The wider the dynamic range the less accurate the output at the ends of the range will be. To solve this problem a switch can be installed that reduces the input with 20 dB. Another option is placing an adjustable gain control in the amplification circuit. In this way the sound sensor may have a smaller dynamic range but it is op-timally used in different situations.
The dynamic range of most integrators is now around 50 dB. If this range could be extended, it would not be necessary to split the circuit into a part for low and high signal levels to extend the dynamic range. This would result in less components and thus in a lower cost and power consumption.
REFERENCES
[1] M. Basner, W. Babisch, A. Davis, M. Brink, C. Clark, S. Janssen, and S. Stansfeld, “Auditory and non-auditory effects of noise on health,” The lancet, vol. 383, no. 9925, pp. 1325–1332, 2014.
[2] A. N. S. Institute, American National Standard Specification for Sound Level Meters: ANSI S1. 4A-1985 Amendment to ANSI S1. 4-1983. Acousti-cal Society of America, 1985.
[3] J. P. Bello, C. Mydlarz, and J. Salamon, Sound Analysis in Smart Cities, pp. 373–397. Cham: Springer International Publishing, 2018.
[4] J.-A. Fernandez-Prieto, J. Ca˜nada-Bago, and M.-A. Gadeo-Martos, “Wire-less acoustic sensor nodes for noise monitoring in the city of linares (ja´en),” Sensors, vol. 20, p. 124, Dec 2019.
[5] I. E. Commission et al., “Electroacoustics—sound level meters—part 1: Specifications (iec 61672-1),” Geneva, Switzerland, 2013.
[6] T. Van Renterghem, P. Thomas, F. Dominguez, S. Dauwe, A. Touhafi, B. Dhoedt, and D. Botteldooren, “On the ability of consumer electronics microphones for environmental noise monitoring,” Journal of Environmen-tal Monitoring, vol. 13, no. 3, pp. 544–552, 2011.
[7] A. E.-L.-L. Contours, “International standard iso 226: 2003,” International Organization for Standardization, Geneva, Switzerland, 2003.
[8] Uwe Beis / Rod Elliott, “Wheighting filter set.” http://www.beis. eu/Elektronik/AudioMeasure/WeightingFilters.html. [9] S. M¨uller, “Transfer-function measurement with sweeps director ’ s cut
Contents
List of Figures iii
List of Tables v
List of Abbreviations vi
1 Introduction 1
1.1 Sound level meter . . . 1
1.1.1 History of the sound level meter . . . 2
1.1.2 Smartphone as sound level meter . . . 5
1.2 Applications options . . . 6
1.2.1 Smart Cities . . . 6
1.2.2 Citizen Science projects . . . 7
1.3 Motivation and goal . . . 8
1.4 Contents of this thesis . . . 9
1.5 Impact of the corona crisis . . . 9
2 Specifications 10 2.1 Sound pressure level . . . 10
2.2 Dynamic range . . . 11 2.3 Time weighting . . . 11 2.4 Frequency weighting . . . 12 2.4.1 A-Weighting . . . 12 2.4.2 C-Weighting . . . 13 2.4.3 Octave-band filters . . . 14 2.5 Standards . . . 15 2.5.1 ANSI . . . 15 2.5.2 IEC . . . 16 2.6 Power . . . 16 3 Design 17 3.1 Microphone FG-23329-P07 . . . 17 3.2 Amplification . . . 18 3.3 Frequency filtering . . . 20
3.4 Integration . . . 22 3.5 Processing of output . . . 24 4 Implementation 26 4.1 Spice simulations . . . 26 4.2 Design in Eagle . . . 26 4.2.1 PCB fabrication . . . 27 5 Results 30 5.1 Simulation results . . . 30 5.1.1 Amplification . . . 30 5.1.2 A-weighting circuit . . . 31 5.1.3 Integrator circuit . . . 34 5.2 Measurement results . . . 41 5.2.1 Set Up . . . 41 5.2.2 Amplification . . . 43 5.2.3 A-weighting circuit . . . 44
5.3 Overview complete signal flow . . . 48
5.4 Operation conditions . . . 49
5.5 Power . . . 49
5.6 Cost . . . 50
6 Conclusion 52 6.1 Improvements to the design . . . 52
6.2 Further research . . . 53
A Matlab code 54
B Schematic of circuit 55
List of Figures
1.1 Type 1551A Sound level meter. . . 4
2.1 Equal Loudness Curves curves. . . 13
2.2 Standard weighting curves. . . 14
3.1 Schematic representation of the design. . . 17
3.2 Schematic representation of the design with division of high and low signal levels. 19 3.3 Amplification circuit for low signal levels. . . 19
3.4 Amplification circuit for high signal levels. . . 20
3.5 A-weighting circuit. . . 21
3.6 Transferfunction of A-weighting circuit. . . 22
3.7 Integrator circuit. . . 24
4.1 Top layer of the PCB design. . . 27
4.2 Bottom layer of the PCB design. . . 28
4.3 Components soldered on the PCB. . . 28
4.4 Capacitors in parallel. . . 29
5.1 Amplification of the signal. . . 30
5.2 Simulation results of a linear sweep of the A-weighting circuit. . . 31
5.3 Standard A-weighting curve and transfer function from linear sweep simulation. . 32
5.4 Method to recover impulse response, using an exponential sweep. . . 32
5.5 Input and output exponential sweep in spice simulations. . . 33
5.6 Comparison of transfer function obtained by using an exponential sweep and linear sweep in Spice. . . 34
5.7 Output of integrator at calibration level. . . 35
5.8 Output of integrator at calibration level with fast time weighting. . . 35
5.9 Output of integrator at calibration level with slow time weighting. . . 36
5.10 Tukey window. . . 36
5.11 Burst signals with different duration. . . 37
5.12 Output of integrator with fast weighting for burst signals of different duration. . 38
5.13 Output of integrator with slow weighting for burst signals of different duration. . 38
5.14 dB output with sine (A=123.7 mV rms, f=100 Hz) as input. . . 39
5.16 dB output with sine (A=123.7mV rms, f=10 kHz) as input. . . 40 5.17 Set up of the measurement. . . 43 5.18 Picture of amplification of high signal levels. . . 44 5.19 The measurement results of the time signal of the input and output of the
A-weighting circuit. . . 44 5.20 The transfer function of the weighting circuit compared to the standard
A-weighting curve. . . 45 5.21 Simulation and measurement results of output integrating circuit. . . 47
List of Tables
3.1 Overview of the poles in the UPM 550 circuit and ANSI standard. . . 22
3.2 Comparison of different types of integrators. . . 23
4.1 Packages of components used on this PCB. . . 27
5.1 Comparison between standard A-weighting and linear sweep simulation. . . 33
5.2 Comparison of the deviations in output between burst signals of different duration and a continuous signal. . . 39
5.3 Mean value of output signal integrator for different frequencies and different time weightings. . . 40
5.4 Ripple on output signal integrator for different frequencies and different time weightings. . . 41
5.5 Simulation results of output from integrator, with varying amplitude of sinusoidal signal at 1kHz as input. . . 42
5.6 Comparison between standard A-weighting and exponential sweep measurements. 46 5.7 Output of integrator for sinusoidal signals from 10mV rms to 400mV rms. . . 47
5.8 Overview of signal flow. . . 48
5.9 Overview of power consumption. . . 50
List of Abbreviations
AC Alternating Current
ANSI American National Standards Institute ASA Acoustical Society of America
CDIP-SB Side-Braze Ceramic Dual In-Line Package DAW Digital Audio Workstation
DC Direct Current
FR4 Flame Retardant 4
GPS global positioning system
IEC International Electrotechnical Commission IoT Internet-of-Things
ISO International Organization for Standardiza-tion
ITU-T International Telecommunication Union -Telecommunication
Leq equivalent continuous sound level
MCS Mobile Crowd Sensing
MEMS micro-electro-mechanical-system PCB printed circuit board
rms root-mean-square
SO Small Outline
SOIC small outline integrated circuit SONYC Sounds of New York City SPL Sound Pressure Level
VSSOP Very Thin Shrink Small Outline Package WASN wireless acoustic sensor networks
WAV Waveform Audio File Format WHO World Health Organisation
Chapter 1
Introduction
Every day people are exposed to unwanted noise. Since the 1960’s scientific studies have shown that exposure to noise can cause damage to the human health. This can be directly related to auditory damage such as hearing loss or tinnitus, but it can also effect non-auditory health damage such as stress, cardiovascular diseases, sleep disturbance and cognitive performance [1]. The non-auditory damage is especially caused by environmental noise. According to the World Health Organisation (WHO) [2], environmental noise is noise emitted from all sources except for noise at the industrial workplace. It is mainly caused by outdoor noise such as transport (rail, road and air traffic), construction and social events (bars, neighbours, fireworks, concerts). The degree of damage is determined by the sound level and the exposure time. To reduce the effects on public health, the noise pollution has to be analysed by studying the results of sound level measurements.
Measuring the sound pressure level is more difficult compared to other quantities, such as air pollution or meteorological conditions, because the sound pressure carries much more informa-tion in the signal and varies faster in time than the other quantities. The challenge is to average the sound pressure over time in an energy efficient way. Therefore sound levels are integrated over time in energy equivalent levels, which is needed to compare measurements with different duration. Only the frequencies audible for the auditory system should be taken into account.
In this chapter a brief explanation of the main concepts of this master thesis is given. The history of the sound level meter, the transition from analog to digital meters and today’s sound level meters are discussed. Next, possible application options such as smart cities and citizen science projects are suggested, taking into account the challenges and trade-offs of designing a sound level meter. This chapter ends with a description of the content and goal of this thesis.
1.1
Sound level meter
A sound level meter or sound pressure level meter is an acoustic instrument that measures sound levels according to certain standards. Most of the times it is a portable instrument that exist of
the combination of a microphone and a processing part.
The microphone converts the pressure signal of the noise or sound to an electrical signal. This signal is first amplified and filtered. The filtering of the signal is crucial because the sound level meter should give a result that is comparable with how the human ear would perceive the sound levels. The output of the sound level meter is expressed in decibels.
1.1.1 History of the sound level meter
According to H.H.Scott [3], the New York Noise Abatement Commission were the first to mea-sure a sound level in 1929. They did not use an actual sound level meter but they used the ear-balance method. In this method an audiometer tone is adjusted so that the noise of the environment is equally perceived by the ear as the audiometer tone. This method is not so objective so there was a need to have a standardized instrument that measures the sound level without using the human ear. In 1932 the Acoustical Society of America (ASA) [4] started with the formation of a standard and in 1936 they made a first draft standard (Z24.3 1936). This standard describes a common reference level but in reality there was an uncertainty of 6 dB for the measured levels for a given sound [3].
Analog sound level meter
The sound level meters in this period were heavy, large and expensive. This was mainly caused by the vacuum tube amplifiers and large batteries. The technology evolved very fast, which improved the stability and calibration of the sound level meters. In the 1950’s the sound level meters could reach an accuracy of 1dB on the sound level measurements [3]. The sound level meter without the microphone was evolved really quick as there were better circuit components and tubes available. Hence, the inner circuit did not cause big errors on the measurements, but the microphone was the main cause of the error on the sound level measurements.
The different types of sound level meters were combined with different types of microphones such as carbon, condenser, ribbon or piezoelectric microphones. Each type of microphone had to meet different requirements regarding sensitivity, frequency response, directionality, noise level and stability. Often it was a trade-off to choose the right microphone because of conflicting requirements. A large microphone had the advantage to have a higher sensitivity, which results in lower noise levels. A smaller microphone will have a better directivity and flatter frequency response at higher frequencies. The output of the microphone was also influenced by humidity. The magnetic type of microphone was more resistant to differences in humidity than piezoelec-tric and capacitor microphones, but the piezoelecpiezoelec-tric microphones on the other hand did not require a preamplifier [5].
In the 1950’s most of sound level meters were using a piezoelectric microphone. A crystal microphone produces a voltage that is linked to the deformation of a certain material. In the case of the Shure 98-98 microphone, Rochelle salt was used as piezoelectric material because of
its high output levels and low price. However it was a fragile material and sensitive for defor-mation which results in non-linearity [6]. Therefore other ceramic materials such as barium and lead zicronium titanate were used as well for the crystal microphones. These materials are more stable and reliable than Rochelle salt [5].
A common combination was the Shure 98-98 Rochelle Salt crystal microphone with a Type 1551-A sound level meter. This sound level meter was produced by the company General Radio and it was the successor of the Type 759-A (1936) and Type 759-B (1940). The Type 1551 had many advantages compared to its predecessors and it met the ASA Z24.3 1944 standard (the follow up of the ASA Z24.3 1936). It has a broader frequency-response characteristic, a better stability, a wider dynamic range, a lower noise level, it was smaller, lighter and easier in use [7]. The amplifier circuit of this meter consists of sub-miniature tubes which give a flat frequency response from 20 Hz to 20 kHz. It was also possible to apply standard A, B and C weighting on the measurements. These frequency weightings represent the measurements as it would be perceived by the human ear, because not all the frequencies are perceived equally loud by the ear. Due to the Type CK512AX tube that is used in this sound level meter, it was possible to measure sound levels of 24 dB and the total dynamic range of this meter was thus extended from 24 to 140 dB. The meter was provided by power using two D-size flashlight cells and one portable radio B battery. The operation time of this meter was limited to 20 days, if it was measuring 8 hours a day. If there was a need to use the meter for a longer time on the same place, then it was possible to connect an a-c power supply. The meter was not really sensitive for temperature changes. However if there was a long cable present between the meter and the microphone, the input capacitance of the meter changes and this negatively influenced the reaction on temperature differences. If it was necessary to have more precious results at higher frequencies, another microphone could be connected such as a condenser or ribbon microphone. Due to its good characteristics the 1551 sound level meter was used for noise measurements in the industry, schools and laboratories [7].
In 1961 the American National Standards Institute (ANSI) updated the latest standard to S1.4 1961 and also the International Electrotechnical Commission (IEC) published their first standard for sound level meters: IEC 60123. In 1962 the first hand-held sound level meter that could make very accurate and stable measurements was produced by Br¨uel & Kjær, called Type 2203. It met the requirements of the ANSI and IEC standards. The 2203 sound meter was driven by batteries and weighed only 5 kg, therefore it was ideal for outdoor measurements. It covered a dynamic range of 22 to 134 dB and it was possible to apply A, B, C weighting and external filtering circuits. The microphone accompanied by the 2203 sound meter was an accurate and long term stable condenser microphone. The microphone was robust for varia-tions in temperature and humidity because a cathode follower was used. The cathode follower was placed between the microphone and the input amplifier of the sound meter. This caused impedance matching between the low input capacitance of the microphone and the high input impedance of the sound meter, which prevented extra loss in sensitivity by capacity loading [8].
Figure 1.1: Type 1551A Sound level meter.
Digital sound level meter
The technology of signal processing evolved and with the invention of microprocessors the sound level meters became lighter and smaller in the 1970’s because a higher density of transistors was achieved on an integrated circuit. In the 1980’s, the microprocessor was specialized into a digital signal processor. This gave many opportunities for the sound level meter: calculations could be done in parallel, which made it possible to store the data and display the results of analysis in real time. Before the digital sound level meter, it was only possible to average the time with a fixed time constant that was quite small. The output of the sound level meter could not be stored, which made it impossible to get information about a longer time period. Thanks to dig-ital signal processing, integrating sound level meters arose. These sound level meters use linear integrating circuits, which allows to average measurements over a longer period of time. The output of these sound level meters could now be expressed with a new quantity Leq (=equivalent continuous sound level). This level expresses the equivalent sound level that has the same energy as the total energy of the varying sound level over a given time. The formula of this quantity is explained in chapter 2 (section 2.3). The full history of the measurements could be tracked by using different quantities: peak measurements, Lmax, Lmin, short time integrations,. . .
In the meantime the standards have also been updated to verify the evolutions in the new models of sound level meters. Before 2003 there existed different standards for integrating-averaging sound level meters and sound level meters without integration. The ANSI S1.40,
updated in 2006, and IEC 60942, updated in 2003, are covering specifications for sound calibra-tors. The standards ANSI S1.11, updated in 2004, and the IEC 61260, updated in 2014, define the specifications for octave-band, fractional-octave-band and other fractional filters for both analog and digital sound level meters. The most recent standards IEC 61672 and ANSI S1.4 categorize sound level meters in classes respectively types and determine the specifications that a sound level meter should fulfil to belong to a certain class or type.
1.1.2 Smartphone as sound level meter
Nowadays smartphones or computers can also be used as sound level meters. This makes it more approachable for individuals to measure sound levels or noise. The smartphone as sound level meter has the advantage that it is a hand-held instrument, it is connected to the Wi-Fi and is easily available. However, it also has the disadvantage that the quality of the mobile phones used as sound level meter does not always meet the given standards to have accurate measurements, which is why they are usually not used for professional purposes.
In studies a distinction is made between using the built-in microphone or an external mi-crophone and between iOS or android phones. Among the android phones there are also a lot of differences in model and version, so it is hard to make a general conclusion for all android phones. The iOS mobile phones use all the same audio architecture, which limits the differences in iOS phones and makes it easier to develop a good working application for iOS phones [9].
In the study by Kardous and Shaw [10] iOS phones were tested in a reverberant chamber with as test signal pink noise, that has a dynamic range of 65 to 95 dB and a frequency range of 20 Hz to 20 kHz. They compared the results for each application but none of them met the requirements of the standards, ANSI S1.4 and IEC 61672. The four applications that scored the best were SoundMeter, NoiSee, SPLnFFT and Hunter, they had a deviation of max± 2dB. This study was done a second time [11] but instead of using the built-in microphone of the smart-phones an external microphone, Dayton Audio iMM-6 or the MicW i436, was connected to the devices. The external microphone could be calibrated before the measurements which resulted in more accurate measurements and deviations of only ± 1dB to the reference level. From this study it can be decided that the build-in microphone is the weakest part of the smartphone as sound level meter and not the applications that are running on the phone. The built-in micro-phone is directed to the mouth of the speaker, which means that it is not an omnidirectional microphone. To obtain sound level measurements that are not influenced by the orientation of the device an omnidirectional microphone is needed.
The built-in microphone is mostly a MEMS (micro-electro-mechanical-system) microphone which has a low power consumption, a signal to noise ratio bigger than 60 dB and it can with-stand high temperatures. The MEMS are optimized for speech communication so the frequency response is not always optimally designed to measure other sounds or noise, which is necessary for sound level meters. In another study (in 2018) by Celestina et al. [9] a similar set up is used:
iOS smartphone (iPhone 6), an external calibrated microphone MicW i436 and the NoiSee ap-plication. The measurements between 63 Hz and 8000 kHz meet the requirements of the class 2 of the IEC 61672 standard, which makes it possible to use this set up as an accurate measuring instrument.
1.2
Applications options
A sound level meter can be used to do acoustical measurements. It can test if the noise of certain devices or work environments remains within certain standards. Sound level meters are also used at festivals, concert halls or bars to control the level of the music or they are used to register noise disturbance. If measurements on a larger scale are necessary, for example for smart city applications or citizen science projects, more sound level meters are needed. To sup-press the cost scientist are looking for alternatives such as the smartphone as meter instead of a professional sound level meter.
1.2.1 Smart Cities
A high percentage of the population lives in an urban environment, this can cause an impact on the systems and infrastructure of the city which could result in a decrease of the quality of human live. Therefore technology comes with the concept smart cities. The ITU-T Focus Group [12] defines a smart city as “an innovative city that uses ICT’s and other means to improve quality of live, efficiency of urban operations and services, and competitiveness, while ensuring that it meets the needs of present and future generations with respect to economic social and environmental aspects”. The principle of a smart city system is to gather, distribute and analyse the data.
One of the main problems in a city is noise pollution, this noise contains a mix of many sounds such as traffic sounds, machines, human voices, electronic sounds,. . . which cover the full frequency and temporal spectrum. Long term exposure to noise can have negative conse-quences on human health, sleep disturbance, hearing loss, stress and well-being. Turning the city into a smart city that makes a sound analysis of the city could reduce the problem of noise disturbance. This can be realized by using wireless acoustic sensor networks (WASN). Such a network exist of several fixed sound level meters spread over the city. These sensors can deliver very accurate and continuously measurements in time and space. Such a node consists of a mi-crocontroller, a microphone, an amplifying circuit and a circuit to filter the noise. Noise levels can show large locally differences due to the shielding of buildings or sudden events. By placing many sensors in the city and connecting them to a wireless network immediate interventions can be taken if a noise threshold is exceeded.
In a recent study of Fernandez-Prieto et al. [13], they describe a full implementation of a system that is been tested in Linares (Spain). They used nine sensor nodes. Each node consists
of an Arduino Due device with an antenna, an electret microphone and a Maxim MAX4466. Before the nodes are placed in the city they are calibrated in the lab by using a class 2 sound level meter. The differences between the Arduino Due and the sound level meter are lower than 0.2 dBA in the lab. If the Arduino was placed in an urban environment for a full day the dif-ference was about 1 dBA (measured in Laeq). These larger difdif-ferences are due to the fact that the dynamic range of the sensor is limited to 44-105 dBA, which results in wrong measurements at night because then the noise levels are lower than 44 dBA. The nodes are powered by a 12 V electrical plug and the power consumption for the Arduino devices differs between 125 mA and 900 mA, the microphone and amplifier have a power dissipation of 421 mW. The total cost for a node would be about 110 euros. In this implementation the noise is measured every 30 seconds which ensures that the amount of data accumulates quickly, this can only be sustained for a period of a month. Another concern is the security of the network, it should be save for hacker attacks on the server.
The SONYC project [14] also designs a low cost acoustic sensing device, but they use a Raspberry Pi single-board computer in combination with a MEMS microphone and a Wi-Fi antenna. This project uses a digital design of the MEMS microphone which includes an analog to digital converter and a microcontroller that can communicate with the Raspberry Pi. The dynamic range reaches from 32 to 120 dBA, which is broader than the previous example and thus can measure low noise levels as well. The microphone is installed via a metal goose-neck on the aluminium housing of the sensor. The total cost for one sensor node is 75.6 euros, the cost for the assembly and construction not included.
1.2.2 Citizen Science projects
Another way to construct a noise map of the smart city is to use smartphones as sound level sensors, it is called Mobile Crowd Sensing (MCS). This is a low cost solution which can create large-scale noise maps without major technical interventions. Every smartphone has a built-in microphone, GPS, the possibility to install a given application and is connected to the Wi-Fi or mobile network. Besides the citizen can add extra information about the environment, the type of noise or suggestion on how to reduce the noise on a given place. All the information is collected and analysed by city managers. Important data that must be kept is the time and place of the measurement and the model of smartphone. After the data of the measurements is shortly stored on the mobile phone, it is transmitted to a server where further processing hap-pens. There, links can be established between noise levels and traffic at certain places and times. In the study of Zappatore et al. [15], they tested the accuracy by executing the measurements with a class 2 sound level meter and with various models of smartphones. The measurements resulted in average deviations of ±5 dB. From this, it can be decided that smartphones can be used as instruments to get relevant information about where to intervene for reducing noise pollution. A concept where fixed acoustic sensors are combined with the mobile phone sensors is maybe the ideal solution to get a balance between accuracy, scaling and cost.
Another application of citizen science project is the Amsterdam Sounds project. The noise map of Amsterdam is constructed by several sensor nodes that are handed out to certain citi-zens. The sensor node consists of a SPH0645LM4H-B microphone and an Adafruit Feather M0 RFM95 LoRa microcontroller [16]. The code and assembly method are made open source so that every citizen can make a sensor node and can provide additional measurements. The information is send via an Internet-of-Things and the project aims an accuracy of a class 2 sound level meter.
1.3
Motivation and goal
In recent years there is an increasing need to measure quantities such as temperature, air pollu-tion, noise, humidity, . . . to investigate the influences of these factors on the environment and human health. Therefore sensor networks are set up in the cities. For the most quantities the development of energy-efficient sensors is rather straightforward because sampling only takes place every minute. For sound sensor systems it is more complicated to have energy efficient sensors because the sound pressure level has to be averaged over time and it should be adapted to the frequency range of the human ear. Therefore the urban noise sensor requires digital signal processing or complex electronic circuits.
For digital sound level meters a lot of computing power is needed to calculate the frequency weighting by Fourier transformations especially if a high accuracy in the low frequencies is re-quired. For these calculations, a quite powerful processing board is needed such as a Raspberry Pi 3B+. This model requires a 500 mA current capacity and a voltage supply of 5V, which results in a power consumption of 2.5 Watt. The Raspberry Pi models require a steady voltage supply. Therefore, when it is powered by a battery, caution should be spent on the fact that the battery provides a voltage within the required range for the Raspberry Pi model (4.75V -5.25V) [17]. In acoustic sensor networks the sensor should work as long as possible on the battery, which makes a high energy consuming digital sound level meter not the ideal solution. Therefore more research into analog sound pressure level sensors could be done to design a low-power, low-cost and versatile sensor.
In an analog design the frequency weighting and calculation of the integrated sound level are done by an electronic circuit that has a low power consumption. Only a simple microcontroller is then needed to send the measurements to a central server. The power consumption of such a microcontroller, for example a Arduino Nano (95 mW), is much lower and can work longer on a battery. In this case it would also be possible to use a solar panel as provider for the energy of the sensor.
The goal of the master thesis is to design a low-cost and ultra-low power, plug-and-measure analog sound pressure level sensor. The sensor should be used to measure the sound pressure level in certain environments (city, factories, festivals,...). It should be possible to apply time and frequency weighting on the measurement. The data has to be integrated over time, converted to a
value in decibels and then processed on a microcontroller. Via an Internet-of-Things application the information can be send to a central server. The goal is to approach the accuracy of a class 1 sound level meter of the IEC standards.
1.4
Contents of this thesis
In this thesis the design of an analog sound pressure level meter is discussed. In chapter 2 the important terms and concepts are explained. The classification according to the IEC and ANSI standards is discussed: what are the specifications a sound level meter has to meet if it belongs to a certain class or type. Furthermore the important characteristics for the design of this sound level meter are specified. In chapter 3, the different parts of the design of the sound level meter are described: the choice of the microphone, the circuit to amplify the electric signal of the microphone and the circuits for frequency filtering, the chip for integration and how the output is processed.
In chapter 4, the fabrication process is sketched. First the circuits were simulated in LTspice. When the simulations showed the desired results, the PCB design was drawn in Eagle. This design was then fabricated as a prototype and the different components were soldered on the PCB. In this chapter the problems that turned up during the implementation process are also enumerated. The simulations and measurements results for each part of the design are discussed in chapter 5. To finish a conclusion is made in chapter 6: what could have be done different in the design process and what are improvements that can be made to the design. In the appendix the schematics of the circuit can be found as well as the link to the Matlabcode used for this thesis.
1.5
Impact of the corona crisis
From the 18th of March it was no longer possible to work in the labs of the UGhent. At that time the PCB of the design of this thesis was already produced and soldered. The essential measurements of each part of the design had already been done and some errors were detected. If it would have been possible to still access the labs of the UGhent, some measurements could have been redone to check certain assumptions. In addition, some extra specifications of the design could have been tested on the PCB. If there was time enough, a second PCB could have been fabricated and tested with the microphone attached to the circuit. To compensate for these restrictions, extra simulations were done to investigate the other specifications of the design. Suggestions for improvements on the design or on certain measurements, are also discussed in this thesis. In the introduction, more research was spent on other existing projects and models than originally planned.
Chapter 2
Specifications
The design of the sound pressure level sensor should meet some requirements. The requirements that are pursued, are listed up in this chapter. To clarify the requirements, some concepts are first explained such as sound pressure level, dynamic range, sensitivity, frequency weighting and time weighting. The requirements that are imposed for this design are based on the ANSI and IEC standards. The comparison between the results of the design of this master thesis and the standards is made in chapter 5. To compare the requirements according to power and cost, existing designs are used as criterion.
2.1
Sound pressure level
There is a need for an objective quantity to measure the sound level of a noise or sound source. This quantity is called Sound Pressure Level (SPL) and it is defined as 20 times the logarithm of the sound pressure divided by the reference sound pressure. This sound pressure is the root-mean-square (rms) average of the difference in atmospheric pressure caused by the sound wave. The reference pressure of sound in air is 2∗10−5 N
m2. The unit of SPL is decibels and the formula
is given in (2.1), which is based on the ANSI S1.4 standard [18]. Sound pressure levels lower than 40 dB SPL are experienced as very quiet by the human ear, a level of 120 dB is the pain threshold.
Lp = 20∗ log10(
p
pref) with pref = 2∗ 10
−5 N
m2 (2.1)
To measure the sound pressure level, a microphone converts the sound pressure of the sound wave into an electrical signal. An import factor that determines the efficiency of this conversion is the sensitivity of the microphone. This is determined by the analog output voltage relative to the input pressure [19]. Sensitivity is expressed in mVPa or when the logarithm is taken in dB with reference to 94 dB SPL or 74 dB SPL at 1kHz. The formula is given in (2.2). Microphones with a higher sensitivity will be able to capture lower sound levels better but cannot capture very high sound levels.
SensitivitydBV= 20∗ log10(
SensitivitymV Pa
The sensitivity of a microphone depends on the type of microphone. Condenser microphones have a high sensitivity between -42 and -30 dB. The sensitivity for dynamic microphones is lower, typically from -60 to -48 dB.
2.2
Dynamic range
The dynamic range is determined by the microphone and the amplifying circuit after the micro-phone. It is the ratio between the highest sound pressure level and the lowest sound pressure level the system can handle. The determination of the dynamic range is strongly linked with the sensitivity of the microphone. The lower limit of the dynamic range is defined by the self-noise level. This level is the electric signal the microphone produces when no extra sound source is applied, it is the inherent noise of the microphone. This inherent noise is mainly caused by thermal noise and it will be added to each measurement. Therefore sound levels under the lower limit of the dynamic range cannot be detected by the microphone. The formula to calculate the lower limit (in dB or in pressure) of the dynamic range is given in (2.3). The calculation for the corresponding minimum voltage after conversion is done by using formula (2.4).
SPLmin= 20∗ log10( pmin p0 ) with pref = 2∗ 10 −5 N m2 (2.3) Vmin= SensitivitymV Pa ∗ pmin (2.4)
The upper limit of the dynamic range is defined as the sound pressure from where the output signal is distorted. For a dynamic microphone, this limit is reached when the coil moves out of the magnetic field. For a condenser microphone, the distortion is mostly caused by overloading the electric circuit until the preamp of the microphone reaches its clipping point [20].
For the design of this master thesis a dynamic range of 35 dB to 110 dB is aimed.
2.3
Time weighting
Sound levels vary very quickly and it is hard to display real time variations on the display of the sound level meter. Therefore the meter will smoothen out the sudden changes, which is defined as time weighting. A sound level meter can have three types of exponential time weightings: slow, fast and impulse. The formula according to the ANSI standard S1.4 1983 [18] to calculate the exponential- time-average sound pressure level is given in (2.5), the unit is dB. In this formula the logarithm is taken of the squared ratio of a frequency weighted instantaneous time-varying sound pressure and the reference pressure. This ratio is integrated in time and divided by the time constant τ . According to the standards, the time constant can be 35, 125 or 1000 ms which corresponds respectively to impulse (I), fast (F) or slow (S) time weighting.
Lpτ = 10∗ log( 1 τ Z t −∞ p2() (p0)2e −t− τ d) (2.5)
A slow time weighting is used for sound or noise levels that remain approximately the same. Once the sound source is turned off, the sound level meter should at least decrease by 10 dB in 3 seconds. If the sound source is varying more in time, then the fast time weighting should be used, because the sound level meter has to react on more sudden changes in sound levels. When the sound source is turned off, this should result in a decrease of at least 10 dB in 0.5 seconds. The fast time weighting is ideal to measure urban noise. The third type is the impulse exponential-time-weighting. It is similar to the fast and slow time weightings, but there is an additional peak detector present in the circuit between the exponential averaging and the taking of the logarithm. This peak detector has a rise time of 35 ms and a decay time of 1500 ms. This should result in a decrease of the sound level of at least 2.9 dB in 1 second after the sound source is turned off. The impulse weighting is used to measure short, loud and sudden sounds such as explosions or gunshots.
In this master thesis only the fast and slow time weighting are necessary to implement in the design.
2.4
Frequency weighting
The human ear can only perceive frequencies between 20 Hz and 20 kHz, but not all the fre-quencies in this range are perceived with the same loudness. Therefore, it is essential to apply a weighting on the measured signal from the microphone to simulate the effect of the human ear. There are different types of frequency weightings, but the weightings that are mostly used in sound level meters are the A, C and Z weightings. The curves of the standard weightings are given in figure 2.2. The Z-weighting gives a flat frequency response between 20 Hz and 20 kHz, which can be useful to compare measurements with unweighted levels. The A and C weightings are based on the inverses of the equal loudness curves. Each equal loudness curve represents the same loudness level for the ear and it is expressed in Phon: 40 Phon equals 40 dB SPL at 1 kHz. A curve represents for each frequency the sound level of the pure tone, that is required for the human ear to perceive the tone with the same loudness as the pure tone at 1 kHz. The equal loudness curves are given in figure 2.1, they are based on the ISO 226 standards [21]. As can be seen in the curves, the ear is very sensitive between 1 kHz and 7 kHz which is related to resonance effects in the ear canal. At lower and higher frequencies, a much higher sound level of the pure tone is needed to be perceived as equally loud as the pure tone at 1 kHz.
2.4.1 A-Weighting
The curve for A-weighting is given in figure 2.2 The shape of this curve is approximately the inverse of the 40 Phon equal loudness contour (figure 2.1). As can be seen in figure 2.2, the A-weighting curve suppresses quite a lot the lower and higher frequencies. Therefore A-A-weighting is the ideal frequency filter for noise measurements, because it filters out all the low and high frequencies that are very present in urban noise but not strongly perceived by the ear. According to the standards [18] the A-weighting transfer function is calculated by formula 2.6. This formula
Figure 2.1: Equal Loudness Curves curves.
consists of one pole on the real axis at f2 and f3 and two poles in the complex frequency plane
at f1 and f4. The A-weighting should be zero at 1 kHz, therefore the function is multiplied with
a gain factor of√K1K2. The logarithm is taken of the function and multiplied with 20, because
A-weighting is expressed in decibels and the unit is dBa. The formula for A-weighting can also be written as adding up one pole on the real axis at f2 and f3 to the C-weighting formula and
by adjusting the gain.
WA= 20∗ log( √ K1K2f4 (f2+ f21)(f2+ f24) q f2+ f22 q f2+ f23 ) = WC+ 10∗ log( K2f4 (f2+ f22)(f2+ f23)) (2.6) with f1 = 20.598997 Hz f2= 107.65265 Hz f3 = 737.86223 Hz f4= 12194.22 Hz K1 = 2.242881∗ 1016 K2= 1.562339 2.4.2 C-Weighting
The curve for C-weighting is given in figure 2.2. The shape of this curve is approximately the inverse of the 100 Phon equal loudness contour (figure 2.1). The curve suppresses the lower and higher frequencies less than the A-weighting. It has an almost flat response between 100 Hz and 4 kHz. Therefore, the C-weighting is mostly used to measure louder noise or peaks. The formula of the transfer function of C-weighting is given in 2.7. It consists of two poles in the
complex frequency plane at f1 and f4. The function is multiplied with a gain factor of K1 to
obtain that the C-weighting is zero at 1 kHz. WC= 10∗ log(
K1f4
(f2+ f2
1)2(f2+ f24)2
) (2.7)
Figure 2.2: Standard weighting curves.
2.4.3 Octave-band filters
Sometimes more information about a certain part of the frequency spectrum is needed. There-fore, some sound level meters have the possibility to split up the frequency spectrum by using bandpass filters. According to the standards (ANSI S1.11 [22] and IEC 61620 [23]), the most used bandpass filters are octave-band filters and fractional octave-band filters. An octave-band filter is a bandpass filter whose upper limit is twice the lower limit. The octave-band filters are defined by a band number or midband frequency. This midband frequency can be calculated as shown in formula 2.8.
fm= fmin∗
√
2 = f√max
2 (2.8)
To calculate the upper and lower limit of a fractional octave-band filter, formula 2.9 is used. These filters are also classified by band number or midband frequency, which can be calculated by using formula 2.10. fmax fmin = 21b (2.9) fm= fmin∗ q 21b = fmax p 2b1 (2.10) 1/3 ocatave band filters are the most commonly used fractional octave-band filters for sound level meters. They divide the octave band in three smaller bands, which makes it possible to
give more detailed information about a specific part of the frequency spectrum.
The intention for the design of this master thesis was to implement the several possibilities for frequency weighting: A-weighting, C-weighting and octave band filters. They should have been activated by using a switch, but in this thesis only the A-weighting is realized.
2.5
Standards
There exist different models of sound level meters, each in combination with other types of microphones. To obtain universal measurements that can be compared with each other, cer-tain standards are written. These standards describe the functionalities a sound level meter must have and the specifications on the accuracy of the measurements results. The two main organisations that provide these international standards are the American National Standards Institute (ANSI) and the International Electrotechnical Commission (IEC). Most of the coun-tries adopt these standards or translate them into their own national standards, but they remain very similar. Nowadays, the most adopted standard for sound level meters is the IEC standard.
2.5.1 ANSI
The three important ANSI standards for sound level meters and noise measurements are the ANSI S1.4, ANSI S1.11 and the ANSI S1.43. The ANSI S1.4 is the standard that defines the specifications for the performances of sound level meters. The ANSI S1.11 describes the shape of octave-band and fractional-octave-band for both analog and digital filters. In the ANSI S1.43 standard, the specifications for integrating-averaging sound level meters are listed up. A sound level meter that meets the ANSI S1.4 standard [18] must contain a microphone, an amplifier, a frequency weighting circuit, the possibility to average exponential time and a system to read out the measurements. The ANSI S1.4 standard classifies sound level meters into three types, according to accuracy: type 0, type 1, type 2. The selection of the most suitable type of sound level meter depends on the purpose of the measurement. Type 0 is the most accurate meter and can be used as a reference instrument or in laboratories. For measurements outside the labora-tory, in the free field, a type 1 or 2 can be used. If the sound that should be measured contains a lot of spectral components in the high frequencies (above 3 kHz), then it is recommended to use a type 0 or 1. A type 0 or 1 is also required if there are many variations in the temporal characteristics.
For each part of the sound level meter, the ANSI S1.4 standard imposes limits in which the measurement results may deviate. In addition, there are also tolerances for the overall accuracy for broadband noise and steady sinusoidal signals. For a type 1 instrument these overall errors should be within respectively±1.5 dB and ±1.6 dB and for type 2 the error should be for both measurement signals less than±2.3 dB. The requirements of the standards are compared to the results of the design of this thesis in section 5.
2.5.2 IEC
The current IEC standard that is used for sound level meters is the IEC 61672. This standard is very similar to the ANSI S1.4 standard, except for the fact that the sound level meters are divided in two classes according to accuracy: class 1 and class 2. The IEC standard does not provide an overall accuracy but for class 1 a maximal error of± 1.1 dB at 1 kHz is allowed and for class 2 this error amounts±1.4 dB.
The goal is to develop a design that meets most of the requirements of a type 2 (ANSI) or class 1 (IEC) sound level meter.
2.6
Power
The design of the sound pressure level sensor should be a low power and low cost design. The goal is to develop a sensor that consumes less than 0.1 Watt. The power supply should be powered by a battery or a solar panel. In some cases it is also possible to have rechargeable batteries that are charged by the solar panel. The sensor node should be a portable design so it cannot be powered by a fixed voltage source.
There is always a trade-off between power, cost and accuracy. This sensor should be a low cost design, because the expensive and accurate digital sound level meters are not always the most efficient solution. In case you need many sensors, for example in smart city applications, or if highly strict measurements are not necessary, then the low cost and low power plug–and-measure analog design can be a solution.
Chapter 3
Design
The design of the sound level meter for this thesis consists of five major parts, which are also necessary to meet the requirements of the standards: the microphone, an amplification circuit, a frequency weighting circuit, a time weighting circuit and a processing part. To extend the dynamic range, the circuit is split up in two parts: one for the low signal levels and one for the high signal levels. In this chapter each part of the design is described. The full schematic of the design is added in appendix B.
Figure 3.1: Schematic representation of the design.
3.1
Microphone FG-23329-P07
The microphone for this design is chosen based on earlier research of the Waves group [24]. In this paper the characteristics of eight types of microphones are tested in an outdoor environment for half a year. The microphone that is used in the design of this thesis is the FG-23329-P07, which has comparable characteristics as the microphone that turned out to have the best quality-cost ratio in the study of the Waves group [24]. It is an electret condenser microphone which has the same operating principle as a condenser microphone, except that it does need external power for supplying the small preamplifier of the microphone. The microphone has a diameter of 2.57 mm and it is an omnidirectional microphone, which is required for the design of a sound level meter. In the data sheet the frequency range is defined from 100 Hz to 10 kHz. The sensitivity is -53 dB at 1 kHz with reference 74 dB SPL and -33 dB with reference to 94 dB SPL. The possible error on this sensitivity is± 3 dB. The maximal A-weighted noisefloor level at 1 kHz is 30 dB SPL. Based on this data the dynamic range of the microphone can be calculated. Using formula (2.2) the sensitivity in dB SPL is converted to mV/Pa:
SensitivitymV Pa = 10 −53 20 ∗ 1V 0.1Pa = 22 mV Pa (3.1)
