Implementation and Application of FPGA Platform with Digital MEMS

Implementation and Application of FPGA Platform
with Digital MEMS Microphone Array
Dejan Todorović, Iva Salom, Vladimir Čelebić, and Jurij Prezelj
Abstract— A design of an electroacoustic instrument
prototype for localization of a dominant noise source in the living
and working environment is presented in this paper. Presented
prototype is based on the conventional delay-and-sum
beamforming algorithm which is implemented on an FPGA
(Field Programmable Gate Array) platform. Sound signals are
provided by a receiver system, based on a digital MEMS
microphone array. The circular microphone array is oriented in
a horizontal plane, enabling 360 degrees of acoustic viewing
angle. Configuration enables the implementation of the
beamforming algorithm in real time. Additionally, FPGA brings
many benefits in terms of safety, reliability, rapidity, and power
consumption. The system has been designed using rapid
prototyping methodology with Matlab Simulink tools. A
prototype was built, tested and used to validate numerical
simulations. The prototype was also used for field measurements
of heating plants in Novi Sad and Belgrade. Results of
measurements were successfully applied in definition of sound
sources and model, providing the best background for optimal
noise mitigation measures.
Index Terms—microphone array, MEMS, beamforming,
Matlab-Simulink, FPGA, noise control.
BEAMFORMING is a general signal processing technique
used to control directionality of the reception or transmission
of a signal with an array of transducers [1]. Acoustic
beamforming can be regarded as a spatial filter operation for
the data received from a microphone/sensor array. It can be
used to determine location and intensity of the sound source
[1]-[7]. This technique of sound source localization has found
a wide range of applications in many fields, from acoustic
cameras, medical ultrasound devices, military applications,
cataloguing wildlife in rural areas, videoconferencing, home
surveillance, patient care, to localization of noise pollution
sources in urban environments. Examples of military
applications are localization of the sniper’s position in a
counter-sniper system and localization of submarines using
hydrophones [7], [8].
Dejan Todorović is with Dirigent Acoustics, Mažuranićeva 29, 11050
Belgrade, Serbia, (e-mail:
Iva Salom is with the Mihailo Pupin Institute, Volgina 15, 11060
Belgrade, Serbia; (e-mail:
Vladimir Čelebić is with the Mihailo Pupin Institute, Volgina 15, 11060
Belgrade, Serbia; (e-mail:
Jurij Prezelj is with the Faculty of mechanical engineering, University of
Since the most optimal solution assumes that data from all
the microphones to be processed in parallel, the application of
FPGA (Field Programmable Gate Array) technology arises as
the best solution for the implementation of the beamforming
algorithm on a single module, especially due to the possibility
of adding new algorithms and improving the existing, [8][12]. Additionally, FPGAs are bringing many benefits in
terms of safety, reliability, rapidity, and power consumption.
Recently, digital MEMS (Micro Electro Mechanical)
microphones have been introduced and are nowadays found in
most cell-phones, digital cameras, Bluetooth headsets on the
market. Their quality is continuously improving, and preamplifier, signal conditioning and analogue-to-digital
conversion are often integrated on a single chip. This results
in very compact and lightweight designs [12]. The major
drawbacks using FPGA technology are high costs and need
for integration of peripherals such as AD (analogue-to-digital)
and DA (digital-to-analogue) converters. On the other hand, a
digital MEMS microphone connects to an FPGA directly on
the digital level, without converters.
The facts posed above induced an idea to design an acoustic
system for localization of the dominant noise source by
implementation of the conventional delay-and sum
beamforming algorithm on FPGA platform with a sound
receiver system based on digital MEMS microphones. The
realization of the system, presented in the paper, included
hardware and mechanics system design from the scratch,
hardware and signal integrity verification using HypeLynks
environment, algorithm simulations in Matlab, algorithm
implementation design using rapid prototyping methodology
with Matlab Simulink tools, and connection of all parts of the
system in Xilinx ISE environment. Performance evaluation of
the FPGA design is presented in terms of hardware resources
for the chosen Xilinx Spartan- family. Comparison of the
simulation results and the results obtained on real system is
given, as well.
Beamforming is a method of spatial filtering which
differentiates desired signals from noise and interference
based on their location. The simplest beamforming algorithm
is the delay-and-sum beam former which works by
compensating signal delay to each microphone appropriately
before they are combined using an additive operation. The
outcome of this delayed signal summation is reinforcement of
the desired signal, i.e. the signal the signal coming from a
desired direction (array focus) while suppressing signals
coming from all other directions, and the noise in each
Proceedings of 4th International Conference on Electrical, Electronics and Computing Engineering,
IcETRAN 2017, Kladovo, Serbia, June 05-08, ISBN 978-86-7466-692-0
pp. AKI2.2.1-6
plane wave
Phased Sum
+ =>
Fig. 1 The conventional delay-and-sum algorithm in time domain
microphone tends to cancel each other [8], [13].
The basic idea of the conventional delay-and-sum
beamforming algorithm in time domain is presented in Fig. 1.
It is supposed that sound source is far away from the
microphone array (far field), thus the curved wave front
approaches linearity, with respect to microphone aperture, and
the assumption is that all incoming waves are plane [4]. In a
planar microphone array, the direction vector of a far-field
propagating signal with a bearing of θ is defined by the
unitary vector ⃗. The time Δtm that takes this signal to travel
from a microphone in the array to the origin is proportional to
the projection of the microphone m position vector, rm, on ⃗.
The Δtm shift for each microphone is determined by the
position of the microphone in the array and the desired focus
direction of the array:
Δ =


where c is the speed of sound. The input signal on the
microphone m is an attenuated and a delayed version of the
sound source signal. Signals coming from the same direction
as the focus direction will be amplified after the addition of all
shifted outputs [8]:
a m sm (t − Δt m ).
ssnd (t) = ∑j=1
With the assumption of a planar incoming wave, the
attention of signal at the input of each microphone is the
The conventional delay-and-sum method is the most
primitive of all techniques and not surprisingly has the
greatest hardware requirements [3]. Storage and sampling
prerequisites are of high magnitude and fast clock rates must
be maintained to achieve adequate delay precision. However,
it is the least complex algorithm to implement [4]. Despite (or
even because of) its simplicity, delay-sum beamforming is
still commonly used in many applications [10].
The block diagram of the real-time acoustic beamforming
system is presented in Fig. 2.
A. Sound Signal Acquisition
The sound is acquired using digital MEMS microphones.
These microphones integrate an acoustic transducer, a
preamplifier and a sigma-delta converter into a single chip.
The digital interface allows easy interfacing with the FPGA
without utilizing extra components, such as an analogue-todigital converter, which would be needed for analogue
microphones. The small package size of these microphones
allows for easy handling by a common pick and place
machine when assembling the array of the sensor.
Microphones ADMP621 designed by InvenSense, were used,
because of their good wide-band frequency response from 100
Hz up to 16 kHz, their omnidirectional polar response, high
signal-to-noise-ratio (SNR) of 65 dBA and high sensitivity of
-46 dBFS, making it an excellent choice for far field
applications [14]. These microphones need only a clock signal
between 1 MHz and 3 MHz as the input (apart from the
ground and power supply lines). Each microphone requires 4
lines: ground, power, clock and data. The output of two
microphones is a multiplexed PDM (Pulse Density
Modulation) signal on a single data line using a single clock
source. This means that the total bus width and pins used on
the FPGA to interface the microphones is equal to half the
number of microphones.
B. Sigma-Delta Demodulation (PDM to PCM Conversion)
To get the framed PCM data from the PDM bit stream,
decimation filters are usually used in sigma delta AD
converters. A widely adopted approach in this context is using
CIC (Cascaded Integrator-Comb) filters at the first stage of
decimation to reduce the sampling frequency, followed by 2:1
HB (Half Band) low-pass decimation filters and a LP (Low
Pass) FIR filter to take out the high frequency noise
Fig. 2 Real-time acoustic beamforming system block diagram
introduced in the sigma delta modulation process and the
further decimation [15]-[17]. Since the sigma-delta modulator
inside the microphone is of the 4th order, a 5th order CIC
decimator was implemented. The frequency of the PDM data
output from the microphone (which is the clock input to the
microphone) must be a multiple of the final audio output
needed from the system. In the current implementation, a
decimation of 64 was performed; for the output rate of 48
kHz, thus a clock frequency 3.072 MHz to the microphone
needed to be provided [14], [15].
with flexible flat cables, and houses the FPGA platform
board, providing power supply for the system. 5 V power
supply can be external or provided via USB interface.
C. Delay-and-Sum Block
Delay and Sum block includes conventional delay-and sum
algorithm in time domain and an RMS detector for a chosen
time averaging constant (125 ms or 1 s). The outputs of this
block are: polar stream for directivity pattern presentation (in
current implementation 60 measurement angles (6-degree
step) are calculated), dominant direction stream, a preselected
direction stream and a preselected microphone stream, all
three with the output rate of 48 kHz. Data are transferred to a
PC via USB 2.0 interface. The data acquisition is performed
with an application developed in National Instruments
LabVIEW environment. USB communication is controlled
using NIVISA high-level API (Application Programming
Interface) [18].
A. MEMS Microphone PCBs
MEMS microphones are mounted on small and simple
separated PCBs (Printed Board Circuit) that are connected to
the interface board via USB A type connector, as shown in
Fig. 3. In this way, broken microphones could easily be
replaced, and various microphone array patterns, both planar
and space, could be realized.
B. The Motherboard
The motherboard of the system connects interface boards
Fig. 3 3D platform model
C. Interface Boards
Interface block physically connects microphones to the
central block for acoustic data acquisition and data processing.
Interface block includes three different interface module
types, denoted as module A, B and C, as shown in Fig. 3.
Additionally, interface modules contain holes for mounting
into a designed mechanical construction.
D. FPGA Platform Board
For the realization of the main functions of the system a
USB-FPGA module was chosen, because of its compactness,
price, and the number of differential parts that were brought
out to its general-purpose I/O connectors. The module
contains a Xilinx Spartan-6 LX25 FPGA and a USB 2.0
Cypress microcontroller for interfacing to a PC. The module
is simply plugged into the motherboard of the system thus
quick start of the developed VHDL firmware debugging was
enabled. This provided a considerable time savings since the
effort to incorporate a complicated Spartan-6 LX25 on the
motherboard would not have been insignificant. The interface
chip for USB 2.0 communication was integrated on the
module, with software for performing USB transfer between a
data acquisition PC, running LabVIEW, and the XILINX
Spartan-6. The USB-FPGA module placed on the
motherboard is shown in Fig. 4. The approximate FPGA
resource analysis on XILINX Spartan-6 XCS6LX25 is given
in Table I.
code can be generated using Simulink HDL Coder within the
Xilinx addition for the Simulink – System Generator.
Sometimes there is a benefit from mixture approaches and
such approach is chosen in this design. Block diagram is
shown on Fig. 5. The design consists of the following blocks:
1-Algorithm block, 2-Clock generator, 3- MEMS adapter, 4USB adapter. All the blocks, except the algorithm block were
implemented as VHDL components, while the algorithm
block was built as a model in Simulink.
A. XILINX ISE VHDL Environment
The Xilinx ISE software controls all aspects of the design
flow. Through the Project Navigator interface, one can access
of the design entry and design implementation tools. All
components realized with VHDL or translated Simulink
model are synthetized, translated, placed and routed under
Clock generator synthesize main (system) clock from the
incoming clock sourced by USB controller. The clock
generator is realized with Digital Clock Managers (DCMs).
DCMs provide advanced clocking capabilities to Spartan-6
FPGA applications. DCMs optionally multiply or divide the
incoming clock frequency to synthesize a new clock
frequency. DCMs also eliminate clock skew, thereby
improving system performance. Similarly, a DCM optionally
phase shifts the clock output to delay the incoming clock by a
fraction of the clock period. The DCMs integrate directly with
the FPGA’s global low skew clock distribution network.
MEMS adapter provides all clock signals to MEMS
microphones and performs PDM signal acquisition. PDM
signal is passed to the algorithm block. USB adapter collects
all calculated data from the algorithm block, format data
adding header and check sum and sends formatted data to the
USB controller. It receives control messages sent from the
user application on the PC, as well.
B. Development in the Matlab-Simulink Environment
Fig. 4 USB-FPGA module on the motherboard
Block RAM
app. 84%
app. 63%
app. 71%
FPGA systems are usually designed using hardware
description languages (HDLs) such as VHDL or Verilog in
Xilinx ISE environment. On the other hand, when ModelBased Design is used to target FPGAs, systems can be
designed and simulated with MATLAB and Simulink, and
then bit-true cycle-accurate synthesizable VHDL (or Verilog)
The FPGA design of the system was developed in Simulink
System Generator. System Generator enables use of
predefined elements (primitives) of the elementary logical and
arithmetical circuits [19],[20]. The design is formed as block
diagram with interconnections, as it is usually done in the
Simulink. Signals are separated from basic Simulink
environment, with special elements (input and output ports),
so that the design for FPGA is a separate entity. Most of the
system signal processing is realized in a single component in
Simulink - the algorithm block. This block includes two
smaller blocks: PDM to PCM block and Delay and Sum
PDM to PCM block converts input 1-bit PDM signals at
3072 kHz rate to 48 kHz PCM 16-bit signal. PDM to PCM
conversion consists of three parts: a CIC decimation filter
(realized using Xilinx CIC compiler 2.0) followed by two 2:1
HB filters and a LP FIR filter (all three realized using Xilinx
FIR compiler 6.2).
Delay and Sum block first performs delay function of all
input PCM signals using Dual Port RAM blocks. Delay table
Fig. 5 Block diagram of the HDL architecture of the system
Fig. 6 Matlab simulation results (left) and LabVIEW screen shot (right) of white noise source located at 0 degrees; RMS time constant was 125 ms, N = 8
(1 s averaging time)
values are stored in ROM blocks. After the delay and sum of
all the signals, in the RMS detector values of the polar stream
are calculated for the chosen time averaging constant, and the
dominant direction is calculated.
Microphone array, realized as shown in Fig. 3, comprises
33 digital MEMS microphones, arranged in a specific pattern:
4 concentric rings with radii 25 cm, 40 cm, 55 cm, 70 cm, and
number of microphones 3, 3, 24, 3, respectively. To verify
designed hardware solution, measurements with the
developed system were performed in semi-anechoic EMC
chamber, and the results were compared to the algorithm
simulation results in Matlab. The sound source (PC
loudspeaker) was placed 5 m away from the center of the
system. Measurement signal was white noise, generated from
PC sound card. Signals used in Matlab simulation were
processed to correspond to the real signal. Measurement and
simulation results are shown in Fig. 6. In a measurement
result some reflections can be observed, because of improper
acoustic wall treatment in EMC chamber. It was also shown
that side lobe attenuation at ±30 degrees is better than 12 dB.
The device was used to perform noise in-field
measurements. Scope of the work was to apply appropriate
noise mitigation measures. Each heating plant has several
significant sound sources. Field measurements were
performed at a number of different locations near the
residential areas. Along with the acoustic localization
prototype device presented in this paper, a calibrated
instrument was used for measurements. Fig. 7 shows a detail
from one of the measurements. The acoustic localization
device and a calibrated instrument were mounted on a tripod
and connected to a laptop PC. Power supply for the prototype
device was provided via USB. Due to the low power
consumption of the device, measurement could last up to 6
hours without need for charging or additional power supply
(depending on the power consumption of PC). The polar
diagram shows the position of dominant sound source(s) and
relative levels of significant sound sources, including the
reflected sound from neighboring objects. In this way, sound
Fig. 7 Real life measurement
power radiated from the source in any chosen direction can be
exactly calculated, as well as the influence of reflections and
other significant sound sources.
All data collected during the measurements were recorded
and used later for the modeling. In this way, high accurate
models of the sound source and sound propagation were built
and used for planning of noise mitigation measures.
Because of the flexibility of FPGAs, their communications
and functions can be specialized to provide high performance
for realization of many designs. Capability for implementing
highly parallel arithmetic architectures makes the FPGA
ideally suited for creating high-performance custom data path
processors for tasks such as delay-and-sum beamforming
algorithm. The implementation design described in the paper,
based on FPGA platform and digital MEMS microphones,
presents a good solution for various efficient real-time lowpower systems with beamforming algorithm and a high
number of input signals. Prototype was successfully is being
used to perform measurements in commercial projects.
This work received the partial support from the Serbian
Ministry of Education, Science and Technological
Development, project TR32038.
Ganse C., “An introduction to beamforming,” 2003.
Johnson, D.H.; Dudgeon, D.E., Array Signal Processing: Concepts and
Techniques,” Prentice Hall: Upper Saddle River, NJ, USA, 1993.
Mucci, R.A., “A comparison of efficient beamforming algorithms,”
IEEE Transactions on Acoustics, Speech and Signal Processing, Vol.
32, No. 3, pp. 438-557, 1984.
Harders D., “Development and Implementation of a FPGA based digital
beamformer for an ultrasonic imaging system,” thesis for the Degree of
Bachelor of Digital Systems with Honours, School of Computer Science
and Software Engineering at Monash University, November 2003.
Van Veen B.D., Buckley K.M., “Beamforming: a versatile approach to
spatial filtering,” IEEE ASSP magazine Vol. 5, No. 2, pp: 4-24, 1988.
Christensen, J.; Hald, J., Technical Review Beamforming; Technical
Report 1; Bruel & Kjær: Nærum, Denmark, 2004.
Ul M., “History of acoustic beamforming,” Berlin Beamforming
Conference, November 21-22, 2006.
Tiete J, Domínguez F, da Silva B, Segers L, Steenhaut K, Touhafi A.,
“SoundCompass: a distributed MEMS microphone array-based sensor
for sound source localization,” Sensors (Basel). 2014 Jan 23;
Zimmermann, B., Studer, C., “FPGA-based real-time acoustic camera
prototype,” Proceedings of 2010 IEEE International Symposium on
Circuits and Systems (ISCAS), Paris, France, May 30-June 2 2010.
Graham P., Nelson B., “FPGA-Based Sonar Processing,” Proceedings
of the Sixth ACM/SIGDA International Symposium on FieldProgrammable Gate Arrays (FPGA ’98), pp. 201-208, 1998.
Tomov, B.G., Jensen, J.A., “A new architecture for a single-chip multichannel beamformer based on a standard FPGA,” IEEE Ultrasonics
Symposium, Vol. 2, pp. 1529-1533, Atlanta, GA, 07 Oct 2001-10 Oct
Perrodin, F., Nikolic, J., Busset, J., Siegwart, R., “Design and
calibration of large microphone arrays for robotic applications,”
IEEE/RSJ International Conference on Intelligent Robots and Systems
(IROS 2012), pp. 4596-4601, Vilamoura, 7-12 Oct. 2012.
Ye H., Whittington J., Himawan I., Kleinschmidt T., Mason M., “FPGA
implementation of dual-microphone delay-and-sum beamforming for incar speech enhancement and recognition,” AutoCRC Conference 2009,
Conference Proceedings, Melbourne, Convention and Exhibition
Centre, Melbourne, Victoria, 5 March 2009.
InvenSnese: ADMP621 datasheet — Wide Dynamic Range Microphone
with PDM Digital Output; Technical Report DS-ADMP621-00;
InvenSense Inc. San Jose, CA, USA, 2013.
Uchagaonkar P.A., Shinde S.A., Patil V.V., Kamat R.K., “FPGA based
sigma –Delta analogue to digital converter design,” International
Journal of Electronics and Computer Science Engineering 1(2), pp. 508513, 2012.
Hegde N., “Seamlessly Interfacing MEMS Microphones with Blackfin
Processors,” Analog Devices, Engineer-to-Engineer Note EE-350, Rev
1 – August 3, 2010.
Milić Lj., “Multirate Filtering for Digital Signal Processing: MATLAB
Applications,” Information Science Reference, New York, 2009.
USB Instrument Control Tutorial, Jan 16, 2014., National Instruments,
S. T. Karris, “Introduction to Simulink with Engineering Applications,”
2nd ed., Orchard Publications, 2008.
System Generator for DSP Reference Guide, UG638 V11.4, December
2, 2009,
Download PDF
Similar pages