BAM!

Massive Audio Beamforming December 2, 2016

Design Specification
Editor: Klas Gudmundsson
Version 0.1

Status
Reviewed
Approved

TSKS05 1 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016

PROJECT IDENTITY
Group members
Name Responsibility Phone Email
Linnea Faxén Project Manager (PM) 073-829 65 09 [email protected]
Klas Gudmundsson Documentation Manager (DM) 072-714 06 66 [email protected]
Eskil Jörgensen Chief of Design (CoD) 076-222 53 72 [email protected]
Stefan Lundström Hardware Specialist 070-379 80 79 [email protected]
Ema Becirovic Test Manager (TM) 073 564 15 73 [email protected]
Javier Preciado Software Specialist +34 647 92 61 38 [email protected]

Customer: ISY, Linköpings universitet, 581 83 Linköping

Customer contact: Mikael Olofsson, [email protected]
Course leader: Danyo Danev, [email protected]
Tutor: Christopher Mollén, [email protected]

TSKS05 i Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
Contents
Document History iii

1 Introduction 1
1.1 Purpose of BAM! . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Overview of the System 2

3 Hardware 3
3.1 Central Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3.1.1 Computer . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3.1.2 Expansion Box . . . . . . . . . . . . . . . . . . . . . . . . 4
3.1.3 Distribution Box . . . . . . . . . . . . . . . . . . . . . . . 4
3.1.4 Cross-coupling board . . . . . . . . . . . . . . . . . . . . . 5
3.1.5 Digital Boards . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Signal Splitters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.3 L/M Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.4 Power Supply . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4 Software 9

5 API 10
5.1 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5.1.1 Assignment of Arrays and Terminals . . . . . . . . . . . . 11
5.1.2 From MATLAB via USB to Master Arduino . . . . . . . 11
5.1.3 From Master Arduino via I2C to Arduino Slaves . . . . . 12
5.1.4 From MATLAB to L/M units . . . . . . . . . . . . . . . . 12
5.2 Data Acquisition Library . . . . . . . . . . . . . . . . . . . . . . 13
5.2.1 DAL Function "init" . . . . . . . . . . . . . . . . . . . . 14
5.2.2 DAL Function "transmit" . . . . . . . . . . . . . . . . . 14
5.2.3 DAL Function "record" . . . . . . . . . . . . . . . . . . . 14
5.2.4 Running DAL from MATLAB . . . . . . . . . . . . . . . 14
5.2.5 Risks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

6 Application 14
6.1 Background theory . . . . . . . . . . . . . . . . . . . . . . . . . . 15
6.1.1 Mathematical notation . . . . . . . . . . . . . . . . . . . . 15
6.1.2 Linear beamforming precoders . . . . . . . . . . . . . . . 16
6.1.3 Channel Estimation . . . . . . . . . . . . . . . . . . . . . 18
6.1.4 Equalization . . . . . . . . . . . . . . . . . . . . . . . . . 18
6.2 Directive Sound Transmission . . . . . . . . . . . . . . . . . . . . 18
6.2.1 Pilot generator . . . . . . . . . . . . . . . . . . . . . . . . 19
6.2.2 Channel estimator . . . . . . . . . . . . . . . . . . . . . . 20
6.2.3 Precoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
6.2.4 Main program . . . . . . . . . . . . . . . . . . . . . . . . 21
6.3 User Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

References 23

TSKS05 ii Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
Document History
Version Date Changes Sign Reviewed
0.1 2016-09-30 First draft. KG EJ,EB,KG

TSKS05 iii Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
1 Introduction
In the course TSKS05, a group of students are tasked with a project. This year,
the group specified under Project Identity have been given the task to construct
a Massive Audio Beamformer (MAB) that can be used as a demonstrator for
Massive Multiple Input, Multiple Output (MIMO). This document will provide
a detailed description of the planned system so that the reader can get to know
how it will work.
The MAB will be able to focus, or beamform, sound to terminals. It will
have an array that transmits signals. By beamforming it will direct these signals
to specific terminals. Terminals will receive the signal intended for them along
with noise but should not receive the signals of the other terminals. The MAB
is illustrated in Figure 1. The group has decided to name this Massive Audio
Beamformer ”BAM!” which is MAB backwards and then an exclamation mark
at the end to make it sound more powerful.

Figure 1: Example of finished product

A user will interact with the MAB via a user interface on a computer. Us-
ing the MAB the user can control which antennas should be in the array or
in the terminals and what data to transmit. The antennas in the MAB will
be 64 combined loudspeaker and microphone units (L/M units). Since the
MAB will beamform audio, transmitting a signal corresponds to playing it in
a loudspeaker. Receiving a signal corresponds to recording the sound, using a
microphone.
To demonstrate the powers of massive MIMO in a pedagogic manner, two
different sounds will be played at two different terminals in a room. The sounds
are played from the array but through beamforming only one of the sounds is
heard at one terminal and the other sound at the other terminal. This will
demonstrate the abilities of massive MIMO to direct signals to different users.

1.1 Purpose of BAM!

The purpose of BAM! is to be an easy to use, easy to upgrade, demonstrator
for massive MIMO.

TSKS05 1 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
1.2 Definitions
Throughout this document a number of abbreviations are used. All of them are
defined in Table 1.

Table 1: Abbreviations used in the document

Abbreviation Definition
A/D Analog to digital
API Application programming interface
APP MAB Application
D/A Digital to analog
DAL Data Acquisition Library
GUI Graphical user interface
I2C Inter-Integrated Circuit
L/M unit Loudspeaker/microphone unit
LED Light-emitting diode
MAB Massive Audio Beamformer
MIMO Multiple input multiple output
PCI Peripheral Component Interconnect
SNR Signal to Noise Ratio
USB Universal Serial Bus

2 Overview of the System

The system consists of two subsystems, hardware and software. Software is
further divided into an application programming interface (API) and an Appli-
cation. The user interacts with the Application which uses functions from the
API to control the hardware. The hardware contains all L/M units and the
necessary electronics to control them. Figure 2 shows a basic overview of the
system. There are three large areas of the project: assembling the hardware,
making the API work and creating an Application. In the coming sections, the
hardware, API and Application will be described.

Application API
Hardware

Software

Figure 2: System overview

The hardware is explained in detail in Section 3.

TSKS05 2 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
3 Hardware
The design of the whole system hardware has been created by the customer.
What remains is to construct the hardware according to the design and control
it via a computer. To construct the hardware all L/M units must be soldered,
assembled and tested. Some other parts of the hardware described in the coming
sections such as signal splitters and cross coupling board must also be assembled
and tested. In order to control the hardware, the data to transmit must be
converted from digital to analog, the data to receive must be converted from
analog to digital and the mode of each L/M unit must be set. Figure 3 shows
an overview of the hardware and its subsystems.

Figure 3: Hardware overview

As can be seen in Figure 3, the hardware will have a central unit that directs
the data to send and control signals from the computer to the L/M units with
the help of two expansion boxes and a distribution box. It will also direct the
received data from the L/M units to the computer. The cables carrying the
signals to and from the 64 speakers are connected to the central unit. There are
16 cables, going out from the distribution box which are split into 4 new cables
each at the signal splitters. A power supply will also be located in the central
unit. The power is then split with power splitters and distributed to all L/M
units. The rest of this chapter will describe each part of the hardware.

3.1 Central Unit

The purpose of the central unit is to direct all signals. This consists of directing
the data to transmit to the D/A-card, the converted signals to the correct L/M-
units, directing the received data from the L/M-units to the A/D-converter,
directing the converted signals back to the computer and directing control sig-
nals from the computer. It will consist of a computer, two expansion boxes, a
distribution box and a power supply, as can be seen in Figure 3.

3.1.1 Computer
The computer is a Windows PC. Each of the two expansion boxes will be con-
nected via a DVI cable connected to a PCI-Express card on the motherboard

TSKS05 3 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
of the computer. The computer will perform all signal processing needed to
perform beamforming. It will also control the A/D- and D/A-cards.

3.1.2 Expansion Box

The first expansion box will contain four 16-channel A/D-converters and the
second will contain four 16-channel D/A-converters. The D/A-converters are
used to convert the digital signals from the computer to analog signals to be
played at the L/M units. Similarly, the A/D-converters are used to convert the
recorded analog signals at the L/M units into digital signals to be stored in the
computer.
The A/D-boards have a sample rate of 100 ksamples/s in total. The con-
verter converts one channel at a time. Therefore the sample rate per channel
is 6250 samples/s. The D/A-boards have a sample rate of 100 ksamples/s per
channel since it contains 16 D/A-converters that are run simultaneously. There
are four 16-channel boards giving a total of 64 channels. The A/D-board is
of type Contec AD12-16 (PCI) and the D/A-board is of type Contec DA12-16
(PCI).
To avoid folding or aliasing in the sampled signals, the bandwidth of the
sampled signal must be less than half the sampling frequency, also known as the
Nyquist frequency;
fs
B<
2
.
In our system, fs = 6250 Hz as stated above, which gives us that B < 3125
Hz. This limits the possible signals to transmit. The human hearing range
covers 20-20 000 Hz so there is a significant part of it we cannot transmit in.
However, in telephony the voice frequency covers approximately 300-3000 Hz.
Therefore at least speech should be possible to transmit.

3.1.3 Distribution Box

The distribution box contains electronics to distribute power and signals be-
tween the A/D- and D/A-converters and the L/M units. It also contains elec-
tronics to control the transmit and receive relays in the L/M units. In Figure
4, an overview of the distribution box is presented.
The distribution box has three major groups of components for three dif-
ferent tasks. The first group is connection boards which connect data cables
from the D/A-cards to the L/M units. The second group is analog boards that
connect the data cables from the A/D-cards to the L/M units. The third task
is to direct control signals which is done in a cross coupling board. The con-
trol signals from this board is then connected to digital boards that control the
relays in the L/M-units. The digital boards contain Arduinos to distribute the
relay signal to 16 L/M units each.

TSKS05 4 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016

Overview of distribution box

3 + +5V To Indication
5V

230V
Digital digital on front
3A −
GND boards
panel
Power inlet 3 3 3 12V + +12V +12V

230V
module with fuse
1.2A − To
and mains switch 1 Analog
analog
GND
3 12V + boards

230V
Ground to chassis Analog
1.2A − −12V GND

8x8
37 2x16
DB37 - female
Connection −12V
Rear panel. 37 2x16 board 1 +5V
DB37 - female

330
8x8
To DB37 - female
37 2x16

D/A cards Connection

board 2
Dig GND
37 2x16
DB37 - female
4x5 8x8

10 Digital
board 1
Digital
cross
connection 8x8
board
10 Digital
board 2

8x8 16x16
8 25
DB25 - female
DB25 – male 18 10 Analog
A-A board 1 8 25
DB25 - female
8 25
DB25 - female
DB25 – male 18 10 Analog
A-B board 2 8 25
DB25 - female
8 25
DB25 - female
DB25 – male 18 10 Analog
B-A board 3 8 25
DB25 - female Front panel.
8 25
DB25 - female
18 10
Rear panel. DB25 – male
B-B
Analog
board 4 8 25 To signal
DB25 - female
splitter and
8 25
To 18 10
DB25 - female from there
DB25 – male Analog
A/D cards C-A board 5 8 25
DB25 - female
to
8 25
loudspeaker
DB25 – male 18 10 Analog
DB25 - female
boxes.
C-B board 6 8 25
DB25 - female
8 25
DB25 - female
DB25 – male 18 10 Analog
D-A board 7 8 25
DB25 - female
8 25
DB25 - female
DB25 – male 18 10 Analog
D-B board 8 8 25
DB25 - female
16x1
To chassis as close as possible to the connectors.

Figure 4: Distribution Box

3.1.4 Cross-coupling board

The cross-coupling board is a coupling board in the distribution box. Control
signals from the A/D- and D/A-cards can here be directed to ports on the digital
boards. The cross-coupling board can be seen in Figure 5. The configuration of
connections is flexible. Since it is a coupling-board, the cables are not soldered

TSKS05 5 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
but can be easily moved. There are two tasks that the directed control signals
must accomplish. The first is t synchronize the A/D- and D/A-cards. The
second is to direct a signal to the L/M-units relays.
To synchronize the A/D- and D/A-cards they will all use the same external
clock and external trigger. It is important that all A/D-cards are synchronized
with each other and all D/A-cards are synchronised with each other. Only then
will the samples from all channels be transmitted or received simultaneously.
This is necessary to obtain a reliable channel estimation and to avoid transmit-
ting delayed signals. One Arduino on one digital board will generate a clock
pulse and output it on a pin. This is used as an external clock for all A/D- and
D/A-cards. One A/D-card (A/D master) will use one of its digital output pins
to generate an external trigger signal for the other cards.
The A/D master is also used to generate the ’Select’ signal in the Digital
Boards. This is implemented in the hardware by connecting a ’digital output’
pin on this A/D-card to the ’Select’ signals of each Digital Board.
All of the Arduino boards communicate using an I2C bus with dedicated
pins. These pins are also connected using the cross-coupling board.

Digital board for distribution box

3-pole
10-pole IDC connector to cross connection board
connector
To pin 7 of To pin 14 of 1 2 3 4 5 6 7 8 9 10
all CD4070. all CD4070.
+5V
Select

Dig GND

RX A/C
SDA

TX A/C

NC
SCL

TX B/D

RX B/D

+5V

I2C

Circuit for one half – controls relays for 16 channels (part A or C)

+5V +5V
1 2 ¼ CD4070 ¼ CD4070 8 7

10-pole Select Select 10-pole

BC547 22k MOSI SCK 22k BC547
IDC SS MISO IDC
TX A/C TX VIN
3 connector 4 ¼ CD4070 RX A/C RX GND
¼ CD4070 6 connector 5
RESET RESET
to speaker to speaker
Arduino Micro

Select GND 5V +5V Select

- BC547 22k SCL D2 NC 22k BC547 -
SDA D3 NC
We are D4
D5
A5
A4
We are
5 using 6 ¼ CD4070 D6 A3 ¼ CD4070 4 using 3
D7 A2
8 poles. Select D8 A1 Select 8 poles.
BC547 22k D9 A0 22k BC547
D10 AREF
Ch 0-3 D11
D12 USB
3V3
D13
Ch 12-15
7 8 ¼ CD4070 ¼ CD4070 2 1
Select Select
BC547 22k 22k BC547

1 2 ¼ CD4070 ¼ CD4070 8 7

10-pole Select Select 10-pole

BC547 22k 22k BC547
IDC IDC
3 connector 4 ¼ CD4070 ¼ CD4070 6 connector 5
to speaker to speaker
Select Select
- BC547 22k 22k BC547 -
We are We are
5 using 6 ¼ CD4070 ¼ CD4070 4 using 3
8 poles. Select Select 8 poles.
BC547 22k 22k BC547
Ch 4-7 Ch 8-11
7 8 ¼ CD4070 ¼ CD4070 2 1
Select Select
BC547 22k 22k BC547

Figure 5: Digital Board

TSKS05 6 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
3.1.5 Digital Boards
A schematic of half a digital board is illustrated in Figure 5. In total, there are
two digital boards. Each digital board contains two Arduinos that control the
relay of 16 channels each. One board thus controls 32 channels. Channel 1-16
is called A, channel 17-32 is called B, channel 33-48 is called C and channel
49-64 is called D. The first digital board contains A and B while the second
digital board contains C and D. Figure 5 only shows part A (or C) but the
other half including part B (or D) is almost an identical copy. The purpose of
these boards is to output digital signals that are connected to the relays of the
L/M units and switch the behaviour of the L/M unit between loudspeaker and
microphone. The relay signal to the L/M unit is produced by an XOR between
the ’Select’ signal and a signal describing which group the L/M unit belongs to,
in this section referred to as ’group’ signal.
The group signal is unique for each L/M unit and divides the units into
two groups: array units and terminals respectively. The group signals come
from the Arduino boards, which in turn get this information from the user
application. The user application in the computer sends the layout information
via a USB-cable to the the master Arduino board, which distributes it to the
slave Arduinos using I2C.

3.2 Signal Splitters

To lessen the number of cables, boxes called signal splitters are used between
the distribution box and the L/M units. Each signal splitter distributes audio
signals to and from four L/M units. An overview of the signal splitter can be
seen in Figure 6, where DB25 and DB9 are specific types of the D-subminiature
which is a common type of connector.

TSKS05 7 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016

Signal Splitter
Overview

6 DB9
to speaker

6 DB9
to speaker
DB25
from 24
distribution
box
6 DB9
to speaker

6 DB9
to speaker

Figure 6: Signal splitter

3.3 L/M Units

Each L/M unit has an electronics board that handles power and microphone
amplification of signals, as well as switching between transmit and receive mode.
They have two external LED’s, one that indicates if there is power and one that
indicates if the unit is used as a transmitter or receiver. Table 2 describes the in
and output signals from the L/M unit and the unit itself can be seen in Figure
7. The text "OBS!" in Figure 7 refers to the adjustments done to the printed
board, illustrated with black lines.

Table 2: Input and Output pins on the L/M unit

Pin Description
+12 V, 0V and -12 V Power supply
Rel-, Rel+ Chooses whether the L/M unit transmits or receives sound
In- and In+ Input to the L/M unit when used as a loudspeaker
Out- and Out+ Output from the L/M unit when used as a microphone

TSKS05 8 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016

Loudspeaker – Electronic assembly

OBS!

6 1

9 5

1k2
Eller tvärt om.

Power indication
This is a 4-pole connector. One pole outside of the board.
Send indication

Figure 7: Loudspeaker

3.4 Power Supply

All L/M units are powered by the same power supply unit, which is assembled
by the sponsor. The power supply can be seen in Figure 3 as a part of the
central unit. The power supply has four sockets, each consisting of three banana
connectors: -12 V, 0 V and 12 V. These potentials are then distributed to the
L/M units by a series of power splitters.

4 Software
The software for the system will be used to control all speakers. It consists
of two parts: API and Application. The API is the interface to the hardware
and provides functions to use in the Application. The Application is the user
interface and the signal processing needed to perform beamforming. The API
is intended to be easy to use so that different applications (for example from
other projects) can use the same API but perform different signal processing.
This will make it easy for later projects to implement different signal processing
projects on the MAB. An overview of the software, and what interfaces it has
to the hardware is presented in Figure 8.

TSKS05 9 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016

Computer Arduino
I2C
USB I2C
Arduino Arduino

I2C
Arduino

PCI Express
Distribution
Expansion box box

PCI Express
Expansion box

Figure 8: An overview of the tasks of the software

The following chapters will discuss the planned design for the Application
and the API.

5 API
The purpose of the API is to provide three simple functions to the Application:
initialization, transmission of a signal and reception of a signal. The initial-
ization will select L/M units to use in array and terminal as well as setup the
A/D- and D/A-converters. This setup consists of setting sampling frequency,
sampling clock and triggers. In addition to this, the Application should be able
to specify a signal to transmit at an array or a terminal. The recorded signals
should then be returned to the Application. These functions are summarized in
Table 3.

TSKS05 10 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016

Table 3: API Functions

Function Description Input Output
init_boards Initialize the A/D- and - -
D/A-boards
set_array Sets the specified L/M L/M units -
units to group array used for
array
set_terminal Sets the specified L/M L/M units -
units to group terminal used for
terminal
transmit_array Transmits the specified signal to received
signal on configured array transmit signals from
terminal
transmit_terminal Transmits the specified signal to received sig-
signal on configured ter- transmit nals from ar-
minal ray
record_array Record signals at array for duration of received sig-
a given duration recording nals
record_terminal Record signals at terminal duration of received sig-
for a given duration recording nals

5.1 Implementation
To perform initialization, transmission of a signal and reception of a signal,
the API must communicate with the hardware. This will be done using three
different interfaces which can be seen in Figure 8. The Universal Serial Bus
(USB) interface is used to instruct the master Arduino of which L/M units
that will be used as terminals. The indexes of the L/M-units are then directed
to the affected Slave Arduinos via Inter-Integrated Circuit (I2C). Peripheral
Component Interconnect (PCI) Express is the interface to the expansion boxes
that contain the A/D- and D/A-boards. The following subsections will describe
how each interface will be implemented: MATLAB to Master Arduino, Master
Arduino to Slave Arduinos and MATLAB to L/M units.

5.1.1 Assignment of Arrays and Terminals

The Arduino can assign an L/M unit to either terminal or array. However if
the user does not wish to use all available L/M units there must be a way to
handle unused L/M-units. The Application will keep track of if an L/M unit is
a terminal, array or not used. If it is not used no signals will be sent from the
Application to it. Therefore the hardware only keeps track of which L/M units
that are terminals — the others are assumed to be array.

5.1.2 From MATLAB via USB to Master Arduino

As noted in the previous section, the Arduinos only need to know which L/M
units to use as terminals. This information is transmitted through serial com-

TSKS05 11 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
munication via USB to the master Arduino [2]. Table 4 shows what the code in
MATLAB and the code run on the master Arduino will do.

Table 4: Description of code in MATLAB and on Master Arduino

Function MATLAB Master Arduino
set_terminal Transmit number of termi- Receive indexes and send
nals to use and then transmit commands to affected slaves
the index of the chosen termi-
nals

5.1.3 From Master Arduino via I2C to Arduino Slaves

The master Arduino is connected via I2C to the slaves. Indexes for chosen
terminals are here transmitted from the master to the slaves. Table 5 shows
what the code on the master Arduino and the code run on the slave Arduinos
will do. Arduino supports serial communication and has a specific library for
I2C communication which will be used for this interface [3].

Table 5: Description of code on Master Arduino and on Slave Arduinos

Function Master Arduino Slave Arduino
set_terminal Transmit index to affected Receive indexes and set cor-
slave responding output-port

5.1.4 From MATLAB to L/M units

This interface must perform three tasks: set parameters for the A/D- and D/A-
boards, transmit and receive data, and control the select signal that decides if
the array or terminal should transmit. To perform these tasks, the API must
control the A/D- and D/A-boards. These boards are provided with an API,
called API-AIO(WDM) that can be used to control them. The distributor
of the boards, Contec, also provides a Data Acquisition Library (DAL) called
ML-DAQ. By installing this, one can directly use MATLAB:s Data Acquisition
Toolbox to communicate with the boards. However, ML-DAQ only supports
older MATLAB versions and Contec has no plans on updating it. Therefore, a
new, simpler DAL will be written using the provided API-AIO(WDM). This will
be used instead of the ML-DAQ. The chain of communication is illustrated in
Figure 9. To the left, the figure shows what the communication looks like using
ML-DAQ and the Data Acquisition Toolbox. To the right, the figure shows how
the implementation in this project will work.

TSKS05 12 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016

MATLAB MATLAB
Data Acquisition Toolbox Application

Data Acquisition Library Data Acquisition Library

ML-DAQ DAL

Device Driver Device Driver

API-AIO(WDM) API-AIO(WDM)

Hardware Hardware

Figure 9: Communication chain between boards and MATLAB

5.2 Data Acquisition Library

The DAL will be written in C++ in Visual Studio. It will be a set of functions
to be run from MATLAB. The interaction between the Application and the
DAL is illustrated in Table 6 and 7. Table 7 also describes what functions the
DAL will provide. Note that the input bit for selection (0 or 1) is reversed for
the record functions. This is since they will record the signal, not transmit it
at the specified group (array or terminal).

Table 6: API calls to DAL

Function DAL function Input
init_boards init -
transmit_array transmit array, 1
transmit_terminal transmit array, 0
record_array record 0
record_terminal record 1

Table 7: DAL Functions

Function Description Input Output
init Initializes the boards int, char -
transmit Start D/A conversion on the input array and array, int array
A/D conversion on received signals, set select
to the input int
record Start A/D conversion on received signals, set int array
select to the input int

TSKS05 13 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
5.2.1 DAL Function "init"
The purpose of the initialization function is to initialize the settings of all boards.
The settings are: sampling frequency, memory type, clock and triggers. All
A/D- and D/A-boards will use an external clock and external trigger. One of
the digital outputs on one of the A/D-boards will be used as the external trigger.
The sampling frequency is set to the int given as an input.

5.2.2 DAL Function "transmit"

The transmit function starts D/A- and A/D-conversion. It sets Select by setting
one of the digital outputs on the A/D-board to the input int. This function both
transmits and records data.

5.2.3 DAL Function "record"

The record function starts A/D conversion. It sets Select by setting one of the
digital outputs on the A/D-board to the input int. Note that this function does
not start D/A conversion since it only records signals.

5.2.4 Running DAL from MATLAB

To be able to run the functions implemented in C++, MEX-files will be created.
This allows the Application to use the C++ functions from MATLAB as if they
were built-in functions. Practically, the C++ program is modified to include a
wrapper function that handles the input and output to MATLAB. Thereafter,
a binary MEX-file is created. The function can then be run from MATLAB,
which invokes the MEX-file. [4]

5.2.5 Risks
The original idea was to use ML-DAQ and the Data Acquisition Toolbox. How-
ever, forcing the MAB to use an older version of MATLAB would put limitations
on later projects. The MAB would also be an inflexible system. Therefore the
idea is to write a new DAL which will take more time than learning to use the
existing ML-DAQ. A simple DAL will be implemented to start with to see if it
is feasible to do. Thereafter, the amount of time put into implementing it can
be reviewed. If the implementation takes more than 20 hours, the ML-DAQ
will be used instead, together with the Data Acquisition Toolbox, as depicted
in the left part of Figure 9.

6 Application
The Application (APP) is the software system built on top of the API which
consists of two parts, the first part performs some desired audio beamforming
function and the second part provides a convenient user interface. The MAB
system has a great potential for a wide variety of beamforming APPs. Most
of these possible APPs will consist of some basic signal processing and beam-
forming subroutines that are similar. For this reason, some basic MIMO theory
will be introduced in the following section. The theory section is followed by a
section describing the actual APP of this project: directive sound transmission.

TSKS05 14 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
6.1 Background theory
In a MIMO system, there is one set of input elements and one set of output
elements with a channel describing how the output signals depend on the input
signals. In a Massive MIMO system, at least one of the sets contains a ’massive’
amount of elements, which often means at least a few tens of elements. In
proposed Massive MIMO mobile communication systems, the set containing
the most elements is a base station and the other set consists of mobile units.
In the APP of this project, these sets corresponds to the (base station) array
and the terminals respectively. Figure 10 illustrates how the system could look
with 62 array units and 2 terminal units. The array geometry will affect the
system performance but that is outside the scope of this project. For simplicity,
a rectangular array will be assumed, with dimensions adjusted depending on
where the system will be set up. Other array configuration might be considered
in future applications.

Array ………………… Terminal

Uplink
…………………

Downlink

Figure 10: An overview of a system performing beamforming. Array geometry

may vary.

6.1.1 Mathematical notation

Here follows a definition of the mathematical notation used in this document
when describing MIMO theory. Let M be the total number of antennas in the
array and K be the total number of terminals.

TSKS05 15 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
Let
g11 . . . g1K
 
 .. .. .. 
G= . . .  (1)
1 K
gM . . . gM
be the channel matrix where gkm is the channel impulse response between
terminal k and antenna m. If denoted with a hat, Ĝ, it represents the estimated
channel matrix.
Let
φi = φ1i . . . φN


i (2)
be the pilot signal for terminal i. N is the length of the pilot signal which is
used to estimate the channel.
Let  
φ1
 .. 
Φ= .  (3)
φK
be the pilot matrix.
Let the received matrix at the array in the training phase be:

Y = (GΦ)T + N (4)

where N is a noise matrix.

For the downlink, let

q = q1 ... qK (5)
be one sample of each desired signal at the K terminals.
Let  1
w1 . . . w1M


W =  ... .. ..  (6)

. . 
1 M
wK . . . wK
be the precoder matrix.
This gives one sample of the signals to transmit as:

x = qW (7)

x has dimension 1xM, each value xi is a sample to be transmitted at antenna i.

6.1.2 Linear beamforming precoders

Massive MIMO systems work by using beamforming patterns that are different
from phased arrays in that the phaseshift difference between two adjacent ele-
ments does not have to be constant throughout the array. The array units in
turn do not have to be placed in a regular pattern. Instead the beamforming
patterns take all of the channel matrix into account to produce more complex
patterns that increase spatial multiplexing gains.

TSKS05 16 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016

Figure 11: Linear beamforming, [5]

Linear beamforming is done over flat-fading sub-bands, which leads to the

possibility of describing the channel matrix with complex gains. An example is
illustrated with complex channel gains gkm as in equation (1). Every vector q,
transmitted to the receiver is distributed to the transmitter units by multiplica-
tion with a precoding matrix W, so that x = qW is transmitted. The received
vector q̂ will be
q̂ = xG = qWG (8)
as shown in Figure 11. There are different ways of choosing the precoder matrix.
The most common are summarized in table 8.

Table 8: Precoders
Precoder W Description
Maximum ratio (MR) GH MR of signal energy to given terminal
Zero-forcing (ZF) GH (GGH )−1 Completely inverting channel
Regularized ZF GH (GGH + λI)−1 Compromise of the two above

TSKS05 17 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
6.1.3 Channel Estimation
The previous description of precoding requires complete knowledge of the chan-
nel matrix H. Knowledge about the channel is acquired by measuring the
channel, hence H can in principle never be known exactly. However, good
techniques of channel estimation can produce accurate approximations of the
channel matrix.
The channel is assumed to be a linear system. Hence the channel response
can be found by transmitting predefined pilot sequences over the channel. The
pilots should have their energy spread over all of the spectrum of interest in such
a way that the SNR is relatively constant. The pilots should also be orthogonal.
This provides for accurate channel estimation. If the pilots, Φ, are orthonormal
then the channel can be estimated as

Ĝ = YT ΦH ≈ GΦΦH = G (9)

where Y is the received matrix. Pilots should be sent from the terminals to the
arrays. This will make orthogonality easier to achieve since the terminals are
fewer and thus require fewer pilots.
After estimation, the channel frequency response is quantized and subdivided
into flat-fading sub-bands, each having a complex channel matrix and thus a
precoding matrix.

6.1.4 Equalization
Equalization is methods of dealing with frequency-selective fading or frequency-
selective transmission. This is automatically dealt with if unnormalized ZF pre-
coding (see Table 8) is used. Since different APPs have different requirements,
they might use different equalization schemes. In the APP of this project, the
equalization has to be done at transmitter side since directed sound transmission
should work without a terminal units listening.

6.2 Directive Sound Transmission

The aim of the Application is to play several different sounds to several different
terminals, or people, preferably without mutual overhearing. These sounds
could be anything from narrowband sounds to music. One way to do this would
be, to use knowledge of the geometry of the array. However, to demonstrate
massive MIMO, the Application will estimate the channel using a pilot and
then use this estimate to beamform the signal. The subsystems required to
accomplish this is shown in Figure 12. The sound passes through the precoder
and then gets played at the array L/M units. The precoder is determined by
estimating the channel based on recorded pilot signals. The subsystems are
further described in the following sections.

TSKS05 18 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016

Application
Terminals transmit pilots to array
Pilot generator

Channel Array recieves pilots

estimator

Channel estimate

Sound Directive sound transmition

Precoder

Figure 12: An overview of the Application

6.2.1 Pilot generator

The choice of pilots depend on the channel characteristics. Most important
is that they are orthogonal between terminals and that they provide enough
energy in the transmission band. One simple choice would be to sum a subset
of the Fourier series components corresponding to the pilot duration T :
X
φi (n) = sin(2π(n/N )(i/T )) (10)
i∈Ai

The subsets Ai ⊂ N are disjoint so that the pilots φi are orthogonal.

For fading channels, the best way is to alternate consecutive frequencies
between terminals, in order to properly cover all of the band for all terminals.
This is illustrated in Figure 13. Using coherence bandwidth Bc , a p ∈ N should
be chosen small enough so that the frequency spacing ∆f satisfies

∆f := p/T < Bc . (11)

Notice that p is the rate of which pilot frequencies are sampled from the complete
set of orthogonal frequencies. Otherwise, there will be substantial fading that
is not taken into account.

TSKS05 19 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016

Terminal 1

Terminal 2

Figure 13: Frequency distribution of pilots

Pilots have to be transmitted every time the channel is substantially differ-

ent. For the APP of this project, the hope is that the channel will be approxi-
mately static. The idea is that the movement of people has a small impact on
propagation, and also that the movement of people will be as small as possible
during operation. If the channel is static, only one pilot sequence is needed be-
fore the sound transmission. If experiments show that the channel unavoidably
will change in time, then a new pilot sequence has to be transmitted periodically.
See Figure 14.

Static channel
UL DL
Pilots Sound

Dynamic channel
UL DL UL DL UL DL
Pilots Sound Pilots Sound Pilots Sound

Time

Figure 14: Time plot of pilots and sound transmissions

The pilot generator in the Application will have one function that is de-
scribed in Table 9.

Table 9: Pilot functions

Function Description Input Output
generate_pilot Generates pilots M Φ

6.2.2 Channel estimator

As described in Equation (9), the channel estimation will be straightforward if
pilots are orthogonal. The channel estimator part of the Application will have

TSKS05 20 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
one function that is described in Table 10.

Table 10: Channel estimator functions

Function Description Input Output
estimate_channel Estimates the chan- Y, Φ Ĝ
nel

6.2.3 Precoder
To avoid overhearing, the ZF-precoder should be used. However, complete zero-
forcing is only possible with perfect channel state information. If highly accurate
channel estimation is hard to obtain or if the time coherence due to moving
people is far too small, then ZF-precoder might not be the best choice for our
APP. Therefore, ZF-precoding will be implemented and tested. Should the
scenario prove to be too demanding for ZF-precoding, a new solution will be
implemented using MR-precoding. In this case, the precoder of each subband
needs to be scaled to account for frequency-selective fading. The precoder part
of the Application will have two functions described in Table 11.

Table 11: Precoder functions

Function Description Input Output
generate_precoder Generates pre- Ĝ, mode W
coder, see Table
8
generate_signal Generates signal to q, W x
transmit, see (7)

6.2.4 Main program

Having described the Application and the API, the defined functions can be
used to describe the main program.
1 % init
2 M = 2;
3 init_boards () ;
4 set_array ( [ 3 : 6 4 ] ) ;
5 set_terminal ( [ 1 , 2 ] ) ;
6 % main
7 p i l o t = g e n e r a t e _ p i l o t (M) ;
8 response = transmit_terminal ( p i l o t ) ;
9 estimate = estimate_channel ( response , p i l o t ) ;
10 p r e c o d e r = g e n e r a t e _ p r e c o d e r ( e s t i m a t e , ’ZF ’ ) ;
11 transmit_array ( generate_signal ( si g n a l , precoder ) ) ;

TSKS05 21 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
6.3 User Interface
The User Interface (UI) is a graphical user interface to make the system more
user friendly. In the UI the user can choose what L/M units that will be used
for transmitting or receiving sound. There is also a possibility to choose what
sounds or music to be transmitted to each of the terminals. The UI also has
an option if the user wants to save the recorded sound in the terminals. The
graphical user interface is illustrated in Figure 15.

Set array and Choose sound

Test1.wav
terminal

Specify filename for recorded sound

Run functions
Runs the BAM!. Channel is estimated by
Run BAM! pilots. The chosen sound is beamformed to
the terminals.

Transmits the chosen sound at array

Transmit array Transmit terminal or terminal.

Records sound at array or terminal.

Record array Record terminal Recorded sound is saved in specified
filename if provided.

Figure 15: Overview of the graphical user interface

TSKS05 22 Design Specification

 BAM!
Massive Audio Beamforming December 2, 2016
References
[1] T.Svensson and C.Krysander, in LIPS, Version 1.0, Compendium, LiTH,
2002.
[2] Mathworks, in fscanf (serial) [Online]. Available:
https://se.mathworks.com/help/matlab/ref/serial.fscanf.html

[3] Arduino, in Wire Library [Online]. Available:

https://www.arduino.cc/en/Reference/Wire
[4] Mathworks, in Create C Source MEX File [Online]. Available:
https://se.mathworks.com/help/matlab/matlab_external/standalone-
example.html

[5] Thomas L. Marzetta, Erik G. Larsson, Hong Yang, Hien Quoc Ngo, in
Fundamentals of Massive MIMO, 1st ed. Cambridge University Press, 2016

TSKS05 23 Design Specification