Developing of an ultra low noise bolometer biasing circuit

UPTEC F 16018

Examensarbete 30 hp Juni 2016

Developing of an ultra low noise bolometer biasing circuit

Jonas Viklund

Teknisk- naturvetenskaplig fakultet UTH-enheten

Besöksadress:

Ångströmlaboratoriet Lägerhyddsvägen 1 Hus 4, Plan 0

Postadress:

Box 536 751 21 Uppsala

Telefon:

018 – 471 30 03

Telefax:

018 – 471 30 00

Hemsida:

http://www.teknat.uu.se/student

Abstract

Developing of an ultra low noise bolometer biasing circuit

Jonas Viklund

Noise in electronic circuits can sometimes cause problems. It is especially problematic in for example high sensitive sensors and high end audio and video equipment. In audio and video equipment the noise will make its way into the sound and picture reducing the overall quality. Sensors that are constructed to sense extremely small changes can only pick up changes larger than the noise floor of the circuit. By lowering the noise, sensors can achieve higher accuracy.

This thesis presents an ultra low noise solution of the biasing circuitry to the bolometer used in one of FLIR Systems high end cameras. The bolometer uses different adjustable direct current voltage sources and is extremely sensitive to noise.

The purpose is to improve the picture quality and the thermal measurement resolution. A prototype circuit was constructed and in the end of the thesis a final circuit with successful result will be presented.

Handledare: Anders Wistrand

Popul¨ arvetenskaplig sammanfattning

N¨ ar h¨ og kvalitet ¨ ar viktigt i elektroniska produkter kan brus orsaka problem. Som anv¨ andare ¨ ar detta n˚ agot som m¨ ojligtvis huvudsakligen m¨ arks i ljud- och bildprodukter. Ett brus kan till exempel h¨ oras i bakgrunden om man lyssnar p˚ a musik eller s˚ a kan bilden p˚ a en tv inte uppfattas som skarp p˚ a grund av att bakomliggande algoritmer sl¨ atar ut en vanligen brusig bild. Brus ¨ ar ¨ aven ett problem i v¨ aldigt k¨ ansliga sensorer.

Det h¨ ar examensarbetet utf¨ ordes p˚ a FLIR Systems som tillverkar v¨ armekameror. Rapporten behandlar ett problem FLIR Systems har d¨ ar elektroniskt brus st¨ or bildsensorn i en av deras v¨ armekameror.

Bruset i biaseringskretsen till bildsensorn p˚ averkar b˚ ade bild och nog-

grannhet negativt. Rapporten g˚ ar djupare in p˚ a elektroniskt brus och

presenterar en ny och b¨ attre l˚ agbrusig biaseringskrets som l¨ osning p˚ a

problemet.

Contents

Abbreviations

1 Introduction 1

1.1 Background . . . . 1

1.2 Problem definition . . . . 1

1.2.1 Goals . . . . 1

1.2.2 Delimitation . . . . 2

2 Theory 3 2.1 Noise . . . . 3

2.1.1 1/f-noise (pink noise) . . . . 3

2.1.2 Thermal noise (Johnson-Nyquist noise) . . . . 3

2.1.3 Current noise (shot noise) . . . . 4

2.1.4 Other noise . . . . 5

2.2 Power spectral density . . . . 6

2.3 Components . . . . 6

2.3.1 Resistors . . . . 7

2.3.2 Capacitors . . . . 7

2.3.3 Digital to analog converter . . . . 9

2.3.4 Digital potentiometers . . . . 10

2.4 Filter topology . . . . 10

2.4.1 RC-filter . . . . 10

2.4.2 Sallen-Key . . . . 12

2.4.3 Multiple feedback filter . . . . 14

3 Method 16 3.1 Measuring equipment and software . . . . 16

3.1.1 Stanford research systems SR560 Low-noise preamplifier 16 3.1.2 LabVIEW . . . . 17

3.1.3 NI-USB 6361 . . . . 18

3.1.4 Aardvark I ² C/SPI . . . . 18

3.1.5 LTspice IV . . . . 18

3.2 Measuring and analysis . . . . 19

3.3 Phase 1 - Simulating, Measuring and patching . . . . 20

3.4 Phase 2 - New printed circuit board . . . . 22

4 Results and findings 25

4.1 Results from measuring . . . . 25

4.2 Final circuit . . . . 32

5 Discussion 34

5.1 Suggested improvements . . . . 35

6 Conclusions 36

6.1 Future work . . . . 37

7 Appendix 40

Abbreviations

AC Alternating current.

ADC Analog to digital converter.

BOM Bill of materials.

CAD Computer-aided design.

DAC Digital to analog converter.

DAQ Data acquisition device.

DC Direct current.

DUT Device under test.

ESL Equivalent series inductance.

ESR Equivalent series resistance.

GUI Graphical user interface.

I ² C Inter-integrated circuit.

LP-filter Low-pass filter.

LSB Least significant bitr.

OP-amp Operational amplifier.

PCB Printed circuit board.

PSD Power spectral density.

RMS Root mean square.

SMD Surface mounted device.

SPI Serial peripheral interface.

VI Virtual instrument.

1 Introduction

1.1 Background

In an ever growing electronic world where components get smaller and smaller and the demand for devices with high quality picture and audio never seems to stop increasing, electronic noise is causing problems. The noise makes its way through the system and manifests itself in the picture and audio.

When it comes to video applications, different signal processing techniques and algorithms can be applied to reduce the noise of the picture, an example of this is averaging over two or more frames or a median filter. The price to pay is the degradation of image quality and the increase of processing power and energy consumption. Now a day, when a lot of the new devices coming out on the market is battery powered, energy consumption plays a big role when designing new products. The possibility of both lowering the power consumption and improving the image quality by investigating and optimizing the noise of the electronic circuits is something worth exploring.

1.2 Problem definition

Today FLIR Systems is using a too noisy circuit design for the biasing voltages to the bolometer in one of their products. This can be seen in the picture quality and requires them to use a lot of filtering and processing to get a nice, and to the user noise free, picture.

Due to the fact that the existing circuit is a company secret only measure- ments will be compared in this thesis and no schematics of the original circuit will be presented.

1.2.1 Goals

The main focus of this thesis is to come up with a new ultra low noise

design for the biasing circuitry. The circuit provides multiple variable power

supplies to the bolometer which is extremely sensitive to noise.

theory, measurements, component specification and simulations. In the end a new prototype circuit should be produced. This should also be done with respect to a reasonable prize, size, overall producibility and with an energy consumption close to the that of the existing circuit.

1.2.2 Delimitation

The thesis aim to come up with a new circuit and compare the noise perfor-

mance from measurements done on both circuits. The new circuit will not

be tested in an actual IR camera because this would require major rework

of other parts which would not fit into the time schedule. This means that

the actual difference in performance will not be known.

2 Theory

2.1 Noise

In electronic circuits a number of different noise sources is superimposed to form the total noise. Since the noise is stochastic there is no correlation with other noise (for some noises like 1/f noise there can be a small correlation), the total noise is calculated as in equation 2.1 where V _N is the total noise voltage and V _ni is individual noise voltages.

V N = q

V _n1 ² + V _n2 ² ... (2.1) In a lot of applications noise is not a big problem, but when dealing with sensitive sensors, high speed circuits or audio equipment it could cause un- predictable behavior or limit the quality of the product. The sources of these noises will be presented in this section.

2.1.1 1/f-noise (pink noise)

1/f-noise is named after its inverse dependency on the frequency, the lower the frequency, the higher the noise. 1/f-noise has been observed in a lot of things, for example water level, earthquake magnitude and electronics [1].

The origin of this noise is still a bit unclear but when it comes to electronics, some possible reasons are [2]:

•Fluctuations in temperature, which affect the thermal equilibrium

•Migration of impurities

•The resistance changing over time

1/f noise can in some cases be calculated but in practice this requires em- pirically determined parameters [2].

2.1.2 Thermal noise (Johnson-Nyquist noise)

The random movement of free charge carriers in a resistance will result in

Equation 2.2 is the spectral noise power, where S R,th is the spectral noise power, k _b is the Boltzman constant and T is the temperature measured in Kelvin. To calculate the noise voltage and the noise current the spectral noise power is multiplied and divided with resistance respectively. This can be seen in equation 2.3 where R is the resistance [2].

S _R,th = 4k _b T (2.2)

Equation 2.3 is only applicable to ohmic resistance, for impedance the equa- tion needs to be changed, replacing R with the real component of the impedance, Re(Z).

V _Rn,th ² = S _R,th R = 4k _b T R (2.3) i ² _Rn,th = S _R,th

R = 4k b T R

In theory, a capacitor would not create any thermal noise, although in re- ality capacitors have some current leakage and dielectric losses that can be modeled as a loss resistance R _loss in parallel with the capacity C. This is often described as a loss angle δ _C , this angle can be seen in equation 2.4.

tanδ _C = 1

ωCR _loss (2.4)

The noise caused by this is often described as dielectric noise and is calcu- lated with equation 2.5.

i ² _Rn,tδ = 4K _b T ωC

tanδ _C (2.5)

2.1.3 Current noise (shot noise)

Current noise or shot noise is a noise source associated with pn-junctions or

Schottky junctions (although some studies have shown that this can be found

in metallic resistors as well [3]). Current noise is the result of the charge

carriers having to overcome a potential barrier. The different probabilities

of getting to another potential will cause a current.

Equation 2.6 describes the current per time through a cross-sectional area, this can be seen as a random pulse train. N is the total number charge carriers q with δ as the pulse shape function. t _n is the time of emission for the n-th electron.

I(t) =

N

X

n=1

qδ(t − t n ) (2.6)

Each charge will have to get over the potential barrier and some of them will quantum tunnel through, this follows a Poisson distribution. Given that the charges can be described as white noise the current noise is calculated with the equation 2.7 [3].

S(f ) = 2qI (2.7)

Current noise is not as significant as other noises in electronics. Take 1A for example, it consists of 6.24 × 10 ¹⁸ electrons per second. This is such a big number that even if it fluctuate in the billions at any given time this will still be much less current then the 1A itself. It is however temperature and frequency independent which means than in high frequency application and/or at very low temperature, current noise could be the dominating noise source [3].

2.1.4 Other noise

A lot of other factors can cause noise in electronic circuits, one is noise from the main power line getting picked up by the circuit. This is typically showing up as a large spike at and around 50 Hz and the overtones. AM and FM radio as well as air and temperature fluctuations can also induce noise into a system.

There is also the problem of crosstalk. This is when two or more signal traces affect each other by inducing a voltage that can be seen as noise.

This is also a problem in multichannel components such as digital to analog

converters (DACs) where the signal from one channel will ”leak” to another.

with a frequency in the span of human hearing, this noise could manifest itself as auditory noise. This problem goes both ways, meaning that if the PCB is bent or vibrating this would cause the capacitor to generate a small voltage. This is mostly a problem when using surface mounted device (SMD) ceramic capacitors.

A lot of these noise sources can be reduced by a good PCB layout.

2.2 Power spectral density

To analyse noise it is useful to see what frequencies these noise signals con- tain, which is done with spectrum analysing. A time varying signal can be broken down to the individual frequency components regardless of it be- ing an audio signal, radio signals, a simple sine wave or some other signal.

This is often analyzed by computing the power spectral density (PSD), the spectral energy distribution at a given time.

ˆ

x T (ω) = 1

√ T

Z T 0

x(t)e ^−iωt dt. (2.8)

Equation 2.8 computes the Fourier transform of the signal x(t) for the finite interval [0, T ]. The PSD is then defined as equation 2.9 where E is the expected value [4].

S(ω) = lim

T →∞ E[|ˆ x _T (ω)| ² ] (2.9) Depending on the type off signal and its predicted wave content, different windows can be used to improve the effectiveness of the Fourier transform.

The window function is multiplied with the signal x(t) in equation 2.8. When measuring a broad spectrum, as in this case, a uniform window is the best choice [5].

2.3 Components

The formulas for calculating noise in section 2.1 is for an ideal case. Real

components are not ideal and behave differently depending on the type and

manufacturing method. This section will explain some different parameters

of components with respect to noise.

2.3.1 Resistors

Resistors come in many different forms, the most common is through hole mounted and SMDs. In low noise applications SMDs are the better choice due to the fact that the legs of a through hole mounted device can pick up noise, longer legs means more noise. There are different types of SMDs such as thick film, thin film and metal foil. The different types have their pros and cons. Thin film and metal foil have been proven to have better noise characteristics than thick film [6], but is also more expensive. A comparison between the amount of noise in different resistors can be seen in figure 2.1.

Figure 2.1: Shows a comparison of noise in different types of resistors, lower value means lower noise. Picture taken from reference [6]

2.3.2 Capacitors

Figure 2.2 shows the ideal capacitor and the representation of a real capac-

itor.

(a) (b)

Figure 2.2: (a) Show the representation of a ideal capacitor. (b) Shows the representation of a real capacitor.

The real capacitor can be modeled as having a resistor in parallel with the capacitor, this represents the leakage current. It has also got a resistance and a inductance in series with the capacitor. These are called equivalent series resistance (ESR) and equivalent series inductance (ESL). The main difference between different types of capacitors is what the dielectric is made from [7]. Ceramic (multilayered), film and electrolytic is the most commonly used types.

Ceramic capacitors are very common these day because of the low prize, relatively high capacitance to size factor and low ESR/ESL. The capacitance depend on the type of capacitor and what kind of ceramic dielectric that is being used. A ceramic capacitor can have a capacitance ranging from some pF up to a hundred µF . The main problem with the ceramic capacitor is the piezoelectric effect which can translate mechanical vibrations into system noise. Ceramic capacitors which contain a large amount of barium titanate also exhibits a ”pyroelectric” effect (changing due to temperature fluctuations).

Tantalum capacitor is polarized with the anode made of tantalum and

a cathode made from a solid or liquid electrolyte. The dielectric is made

from a very thin oxide and combined with high permittivity this gives the

tantalum capacitor one of the highest capacitance to size of all capacitors.

Solid electrolyte tantalum have a low ESR and have a stable behavior over a large temperature range. A Tantalum capacitor have no piezoelectric like effect but have a higher direct current (DC) leakage current than ceramic and film capacitors. Because of the polarization, the capacitor always have to be connected with the cathode to the the higher potential and the an- ode to the lower potential. Reverse polarization can cause the capacitor to explode violently, which is important to consider when designing a circuit with tantalum capacitors.

Film capacitors have some sort of plastic film as the dielectric, this could for example be polypropylene (PP), polyethylene terephthalate (PET) or polyphenylene sulfide (PPS). They often feature electrodes of aluminum or zinc created by applying the material to the surface of the plastic dielectric.

Film capacitors are constructed like many capacitors in parallel, this gives the film capacitor a very low ESR and ESL, they are also stable and with good temperature characteristics. For a given capacitance a film capacitor is much larger in size than a ceramic or a tantalum capacitor. SMD film capacitors only have a range from pF to some µF . Unlike the ceramic capacitors, film capacitors do not have a piezoelectric effect.

2.3.3 Digital to analog converter

A DAC is a components that converts digital signals to analog signals. They are frequently used in audio equipment as a way of translating digital music into analog music signals. They can also be used to accurately control voltage in a circuit. Some typical parameters of a DAC can be seen in table 2.1.

Table 2.1: Some typical parameters of a DAC.

Parameter Typical Unit

Resolution 8-16 bits

Channels 1-8

Output noise (1-10 Hz) 1-20 µV _p−p

Output noise spectral density 20-150 nV/ √ Hz

Differential nonlinearity (DNL) 0.2-1 Least significant bit (LSB)

Temperature drift 1-25 ppm/ ^◦ C

2.3.4 Digital potentiometers

A digital potentiometer is a component in which the resistance can be digi- tally controlled via some sort of communication protocol, for example serial peripheral interface (SPI) or inter-integrated circuit (I ² C). Some typical pa- rameters of a digital potentiometer can be seen in table 2.2.

Table 2.2: Some typical parameters of a digital potentiometer.

Parameter Typical Unit

Resistance accuracy 10-20 %

Resolution 4-8 bits

Channels 1-2

Output noise (1-10 Hz) Resistor noise µV _p−p Output noise spectral density Resistor noise nV/ √

Hz

Differential nonlinearity (DNL) 0.2-1 Least significant bit (LSB)

Temperature drift 15-50 ppm/ ^◦ C

2.4 Filter topology

Since this thesis primarily focuses on DC voltages, meaning the goal is to filter out as much alternating current (AC) voltages as possible, this section will present some different low-pass filter (LP-filter) designs with their pros and cons.

A LP-filter is a filter that from a certain cut-off frequency will reduce the amplitude of a signal. The filter order is referring to the rate at which the filter reduces the amplitude. In a first order filter the amplitude is reduced by a factor of 2 every time the frequency is doubled and in a second order the factor is 4. This is generalized as f actor = order × 2.

2.4.1 RC-filter

The RC-Filter is a passive first order LP-filter. Figure 2.3 shows a general

representation of a RC-filter.

Figure 2.3: Shows a first order passive RC LP-filter

A capacitor can be seen as a complex impedance that decreases as the fre- quency increases, this can be expressed in equation 2.10 [8]

Z _C

= 1

iωC ₁ (2.10)

The equation for the voltage V _out is the voltage divider equation. Equation 2.11 shows the transfer function from V in to V out , this has a pole in _R ¹

1

C

. H _C

(iω) = V out

V _in = 1

1 + iωC ₁ R ₁ (2.11)

The cut-off frequency can be calculated from equation 2.12.

f c = 1

2πRC (2.12)

The pros and cons of the RC topology in respect to low frequency:

+ Simple design

+ Low component count + Cheap

− Requires relatively large capacitors to get a low cut-off

− No signal amplification

2.4.2 Sallen-Key

Sallen-Key is an active second order filter design which can be used both with a unity and non unity gain configured operational amplifier (OP-amp).

Figure 2.4 shows a unity gain Sallen-Key LP-filter.

Figure 2.4: Shows a second order unity gain Sallen-Key filter

In equation 2.13 the transfer function for figure 2.4 is shown [9]. This can be rewritten as equation 2.14 where ω 0 is the cut-off frequency in radians and is given by 2.15.

H(s) = 1

s ² (R1R2C1C2) + s(R1C1 + R2C1) + 1 (2.13)

H(s) = ω ² ₀

s ² + 2αs + ω ₀ ² (2.14)

ω ₀ = 2πf ₀ = 1

√ R ₁ R ₂ C ₁ C ₂ (2.15)

α is the attenuation coefficient and is given by equation 2.16 where ζ is the

damping factor, or as it is also called, the quality factor Q. The quality factor

describes at which rate the energy of a system is lost. Low quality factor

means big suppression of oscillations and vice versa, this can be calculated from equation 2.17.

2α = 2ω ₀ ζf ₀ = ω ₀ Q = 1

C1

R ₁ + R ₂ R 1 R 2

(2.16)

Q = ω 0

2α =

√ R 1 R 2 C 1 C 2

C2(R ₁ + R ₂ ) (2.17)

As can be seen from equation 2.18 the transfer function has two complex poles and no finite zeros. If these poles have a negative real component the transfer function will be decaying, non-oscillating and stable.

s = −α ± i q

α ² − ω ₀ ² (2.18)

When designing a Sallen-Key filter one will choose the values of the com- ponents based on the wanted damping factor and cut-off frequency. This means that one often chooses the components as in equation 2.19,

R ₁ = mR R ₂ = R C 1 = nC C 2 = C

(2.19)

and equation 2.15 and 2.17 can be rewritten as equation 2.20.

Q =

√ mn

m + 1 (2.20)

ω 0 = 1

RC √ mn

The pros and cons of the Sallen-Key topology in respect to low frequency:

+ Low output impedance

− Requires more components than RC-filter

− Active design, needs a power supply

2.4.3 Multiple feedback filter

Like the Sallen-Key filter, the multiple feedback filter is an active second order filter. Multiple feedback filters is often used when a filter with high gain, high quality factor and/or high frequency functionality is required [10].

A general second order representation can be seen in figure 2.5.

Figure 2.5: Shows a second order general representation of a multiple feed- back filter.

The transfer function is given as equation 2.21 [11], with the variables Q, K and ω 0 expressed in equation 2.22.

H(s) = Kω ₀ ²

s ² + ^ω _Q

s + ω ² ₀ (2.21)

Q =

√ R ₂ R ₃ C ₁ C ₂

C2(R 2 + R 3 + R 2 |K|) (2.22)

ω ₀ = 1

√ R 2 R 3 C 1 C 2

K = − R 3

R ₁

The pros and cons of the multiple feedback topology in respect to low fre- quency:

+ Low output impedance

+ Good with high gain, high Q and high frequency [10]

+ Requires fewer components than Sallen-Key if not in unity gain

− The accuracy of the gain will depend on R ₁ and R ₂ [11]

− Inverting and with a noise gain of factor 2 [10]

− Requires the most components in unity gain

− Active design, needs a power supply

3 Method

This thesis has two different phases. The first one is a phase where mea- surements was taken on a existing PCB with the original circuit to get a better understanding of noise. The PCB was also patched with some new components. The second phase is where a new test PCB with a conceptual new circuit design was constructed. This was done with the components deemed to be the best ones in theory and with respect to the constraints.

3.1 Measuring equipment and software

This section will present the different measuring equipment and software used in this thesis and explains how they were used. In figure 3.1 an overview of the measuring system can be seen.

Figure 3.1: An overview of the system. The computer controls the device under test (DUT) and the preamplifier, a Stanford research systems SR560, amplifies the low noise generated in the DUT. The NI-USB 6361 samples the signal coming from the preamplifier sending it via USB to the computer to later be analysed in LabVIEW

3.1.1 Stanford research systems SR560 Low-noise preamplifier The Stanford research systems SR560 is a widely used low-noise preamplifier used when measuring extremely low noise electrical circuitry. The noise is specified to 4nV / √

hz at 1kHz. It features two configurable filters spanning

from DC to 1 MHz making it possible to construct a high-pass filter, low- pass filter or a bandpass filter. The SR560 will operate with an amplification of 1 to 50000 and on both single ended and differential input signals. To minimize the noise it uses a floating ground and can be totally separated from the main power line by using the built in lead-acid battery [12].

3.1.2 LabVIEW

LabVIEW is a software for system-design and development often used for data acquisition and instrument control purposes. It uses a graphical pro- gramming language named ”G”, which is a dataflow programming language.

This means that the program, or virtual instrument (VI) as they are called, is built up as a block diagram and gives the programmer a good and easy overview of the system. LabVIEW supports multi-processing and multi- threading. The user does not need to schedule any task them self, this is done by the built in scheduler [13].

LabVIEW is built up by VIs and this is done on three different levels, the front panel, block diagram and connector panel. The front panel is the view where the user interact with the VI for example adjusting voltage, viewing the level of a water tank and so on. The block diagram is where the VIs structure is built up and functions defined by connecting different block by virtual connections. These blocks can be for example logical functions as and, or, not, converters from floating point number to a binary vector representation or graphical data viewers. The connector panel is used to create sub-VIs, this means that the programmer can make a new block that can be used in another VI. This is done by connecting inputs and outputs to a block, and the programmer can then specify if it is a port that is required to be connected or optional and even choose a special personalized icon for that block.

With these functions, complex VIs can be built, such as applications to

control a power plant or maybe just a simple measurement rig. These VIs

can then also be built as a standalone application meaning that the end user

will not need their own LabVIEW license to run the application.

3.1.3 NI-USB 6361

The NI-USB 6361 is a high quality data acquisition device (DAQ) from National Instruments. It features in total 16 analog inputs (8 BNC and 8 pin) with a 16 bits resolution and is capable of 2 MS/s when using 1 input and 1 MS/s using multiple inputs. It has, amongst other things, two analog 16 bits outputs with a update rate of 2.86 MS/s and 24 digital I/O lines [14].

Being a product from National Instruments, the NI-USB 6361 interacts nat- urally with LabVIEW making it easy to build custom applications.

3.1.4 Aardvark I ² C/SPI

The Aardvark I ² C/SPI from Total phase is an external host adapter that converts USB to I ² C and SPI [15]. It has drivers for LabVIEW, which makes it possible to communicate with components on a PCB (for example a DAC or an analog to digital converter (ADC)) directly from within a custom application.

3.1.5 LTspice IV

Linear technologies software LTspice (IV) is used to simulate electronic cir-

cuits. The program includes most of Linear technologies own components

but also have a feature to construct your own components and to import

components from other manufacturers. LTspice can simulate a number of

things such as operating points, small signal AC behavior, transient and

noise. These functions may vary depending on how the model component is

constructed [16].

3.2 Measuring and analysis

The preamplifier was AC coupled and set to an amplification of a 1000 times. A low pass filter at 300 kHz was used to remove any unwanted and uninteresting high frequency noise. Then all the measuring was done using the NI USB 636 DAQ and a custom built application in LabVIEW. The applications graphical user interface (GUI) can be seen in figure 3.2. The application is built up by different VIs, some of them are VIs from the NI DAQ driver library. This makes it possible to interact with the NI USB 6361 DAQ, collecting data and setting parameters. A front end GUI is then built up to display the most important data, like the PSD and the root mean square (RMS) value.

Another custom LabVIEW application was built to control the circuits DAC.

This can be seen in figure 3.3. This application communicates with the circuit through the Aardvark USB dongel using SPI.

The data collected with LabVIEW was then exported in a spreadsheet for- mat and the analysing was done using built in functions in MATLAB.

Figure 3.2: Shows the LabVIEW GUI used to collect data.

Figure 3.3: Shows the LabVIEW GUI used to set the DAC voltages.

3.3 Phase 1 - Simulating, Measuring and patching

The first thing done was to get familiar with the original circuit and to analyse it from a theoretical point by looking at the schematics and the datasheets of the components. By doing this some week spots in terms of noise could be found.

Before any measuring was done on the original circuit the noise floor of

the measuring equipment was measured. This was done by putting a 50Ω

termination on the input of the preamplifier. The noise floor acts as a reference on how low noise the equipment was able to measure. The original circuit was then measured to get a control result to later compare with the result from a new circuit. To get good and consistent measurements the DUT was put inside a simple Faraday cage constructed from a small grounded aluminum box. The control result also acted as a foundation for choosing new components and later for designing a new circuit and a new PCB. Parts of new concepts were simulated in LTspice with different components to assess the performance. Figure 3.4 shows a simulation of the ADA4528 OP-amp from Analog devices, a component with very low 1/f noise [17]. After picking out the components and realizing what and how parts of the circuit could be done differently, parts of the original PCB was patched with new components. This was done one by one and the overall performance and component-wise performance was measured.

Figure 3.4: Shows a simulation done in LTspice. It is the noise simulation of a ADA4528 from Analog devices, which features almost no 1/f noise. (The 1/f noise shown in the simulations is almost only due to the thermal noise from two resistors in series with the OP-amp, a total resistance of 56kΩ).

It was quickly established that patching the original PCB introduced a lot

of new noise into the system. Since most of the components patched did not

in figure 3.5, where a DAC and an OP-amp was patched onto the original PCB. The patching did not prove much in terms of the noise performance for the new components but did serve as a proof of concept for phase two.

The result from phase one is presented in section 4.

(a) (b)

Figure 3.5: (a) Showing a ten times magnified picture of a patched DAC.

(b) Showing a ten times magnified picture of a patched OP-amp

3.4 Phase 2 - New printed circuit board

After the findings and from the simulations in phase one, the next step was to come up with a new and better circuit. In the new circuit there were some components that was more or less obligatory in order to meet the project specifications, not including passive components.

•A low dropout regulator was needed to regulate the external voltage, in this case a battery, to a suitable voltage to use in the new circuit.

•A voltage reference, to get a very stable and accurate low noise reference voltage source.

•OP-amps were needed in filters and to amplify the reference voltage to a suitable potential.

Then there was two ways of regulating the voltage, either a DAC or by the

use of a digital potentiometer regulating the gain in the amplification cir- cuit and attenuation. From table 2.1 and 2.2 in section 2 it looks like DACs generally have better temperature characteristics, accuracy and resolution.

Only to correct for the problem with the low resolution of the digital poten- tiometers would require the use of two components/two channels and this also make for more difficult communication. This fact alone might not be a reason not to used digital potentiometers but together with the uncertainty of the resistor accuracy which would make predictability very hard, it was deemed better to use a DAC design.

After figuring out how the new circuit should be designed, the schematics and the layout of a new test PCB was constructed. The test PCB was constructed with the computer-aided design (CAD) software Eagle from CadSoft. Extra footprints were added for decoupling capacitors. This was done so that it would be possible to play around with different values and also to be able to patch in capacitors with larger footprints.

The test PCB can be seen in figure 3.6. It is a two-sided board with compo- nents only on the top side and where all untraced board is filled with ground plane. The full schematics and a bill of materials (BOM) can be seen in the appendix.

Measuring on the new test circuit was done with the same set up as in phase

one and the result is presented in section 4.

Figure 3.6: Shows the layout of the test PCB. The dimensions is in millime-

ter.

4 Results and findings

This section will present all the results from the measuring done on the different circuits and in different configurations. In the end of this section the final design will be presented.

4.1 Results from measuring

In figure 4.1 the results from comparing the noise of a amplifier at different gain can be seen. This was done to confirm the Sallen-Key filter in unity gain with a noise gain of one is better then a multiple feedback filter with a noise gain of two. In this case the comparison was done on the LT1677 from Linear technology in a Sallen-Key filter configuration but with varying gain. It is clear that not having a gain will improve the amount of 1/f-noise.

The noise increases with the same factor as the gain. This confirms that

the Sallen-Key filter with noise gain one is the better choice, since using a

multiple feedback filter would effectively double the noise.

Figure 4.1: Shows how noise depend on the gain of an amplifier. Measure- ment done on a LT1677 amplifier from Linear technology. The large spikes starting a 50 Hz and with overtones is noise coming from the 50 Hz main power line. The Y-axis is plotted linear to emphasize the 1/f noise.

The control voltage reference used in the original circuit was specified as

ultra low noise with a noise voltage of 1.45 µV _p−p in the frequency range

0.1-10 Hz. The voltage reference LTC6655 from Linear technology chosen for

the new test circuit is specified at the much lower 625 nV p−p [18]. Figure 4.2

shows that the LTC6655 performs a lot better than the control, especially at

lower frequency. Figure 4.2 confirms that the LTC6655 is a suitable choice

for the final circuit.

Figure 4.2: Shows a comparison of the noise spectral density between the control voltage reference in the original circuit and the new LTC6655 from Linear technology used in the new test circuit.

Figure 4.3 shows a comparison between three different DACs. As can be seen the new DAC candidates, both from Analog devices, performs better than the control DAC. Both the new candidates have a lot of ringing and noise in the higher frequencies, this is most likely due to an imperfect PCB layout and no output capacitor in close relations to the measuring point.

Disregarding this problem the AD5541 have slightly lower 1/f-noise, which

was expected since the AD5066 have 35 time more noise in the 0.1 to 10 Hz

range according to the datasheet [19] [20]. The noise of both the DACs is so

low that almost all the noise seen in 4.3 is noise from the voltage reference

LTC6655 and the noise floor. This is most likely why the measurements do

not show a 35 times difference in performance. Since the AD5066 is cheaper

to use than the AD5541 and the difference in performance is small, the final

circuit will include the AD5066.

Figure 4.3: Shows a comparison of the noise spectral density between the control data from the DAC in the original circuit and the output of the AD5066 and AD5541, both from Analog devices. The reference voltage to the new DACs is the LTC6655. The large spikes around 50 Hz and overtones are noise coming from the 50 Hz main power line.

In phase one there were two OP-amps that were deemed good enough to try on the test PCB, this was the ADA4528 from Analog devices [17] and the LT1677 from Linear technology [21]. Both have very good noise performance figures in the datasheet, but the ADA4528 really stood out. ADA4528 is a chopper OP-amp, this means that it has a built in circuit that compensates for offset voltages giving the OP-amp an offset voltage of just a couple of µV and also virtually no 1/f-noise. The drawback is that they generally do not have a great bandwidth, but in this case that is not an issue. The good specifications of the ADA4528 meant that this also was used as a voltage reference amplifier in the test circuit. In figure 4.4 the noise performance result of the different OP-amps can be seen.

As predicted the ADA4528 have a lower 1/f-noise than the LT1677, but the main problem with the LT1677 is the big spike in noise around 300 kHz.

The ADA4528 was chosen for the final circuit.

Figure 4.4: Shows a comparison of the noise spectral density between two different OP-amps, ADA4528 from Analog devices and LT1677 from Linear technology. The DAC used was the AD5066 from Analog devices.

Figure 4.5 shows a RMS noise comparison between using a ceramic capacitor and a film capacitor. This is measured on the output from the ADA4528, first with a ceramic capacitor then with a film capacitor on the positive feedback loop of the Sallen-Key filter. The spikes shown on the ceramic capacitor trace is from light taps on the table on which the PCB were lying.

The same procedure was done when using a film capacitor. The result

is quite clear and corresponds with the theory. This confirmed that film

capacitors should be used in the positive feedback loop on the Sallen-Key

in the final design.

Figure 4.5: Shows how ceramic and film capacitors affect the total RMS noise. Test was done on a ADA4528 from Analog devices by switching the capacitor connected between the non inverting input and ground.

An overall noise performance comparison for the whole circuit can be seen in figure 4.6 and 4.7. It is quite clear from the figures that the new test circuits has lower noise, however table 4.1 shows a more exact comparison.

In figure 4.6 high frequency noise can be seen, this is most likely due to a

PCB layout problem.

Figure 4.6: Shows a comparison of the noise spectral density between the

original circuit and the best test circuit configuration.

Figure 4.7: Shows a comparison of the noise spectral density between the original circuit and the best test circuit configuration. The Y-axis is plotted linear to emphasize the 1/f noise.

Table 4.1: Overall comparison of RMS noise performance between the new test circuit and the original circuit.

Frequency range Control circuit New test circuit Improvement factor

1-10 Hz 6.3 µV RM S 0.44 µV RM S 14.2

1-100 Hz 15.9 µV RM S 1.58 µV RM S 10

1-1000 Hz 27.5 µV RM S 8.5 µV RM S 3.2

4.2 Final circuit

From the results in section 4.1 a new and final schematic was constructed and can be seen in the appendix with the BOM.

All of the components used in the final circuit is SMDs, this makes the circuit

easy to mass produce using pick and place robots. This means minimum human interaction, which means cheap, fast and accurate manufacturing.

The energy consumption of the final circuit is in theory calculated to be

≈ 0.1W , this is ≈ 9% more then the original circuit. With the noise perfor- mance increase and since this is within the margin of error for the theoretical values taken from the datasheet, this was deemed acceptable.

When calculating the price for the components needed to produce one of the final circuit the total cost is ≈ 500 SEK. The total price when buying components to one of the original circuit is ≈ 280 SEK (the actual price will vary depending on when and from were the components is purchased.

This estimate is done with prizes from Farnell element14 ). The final circuit is more expensive than the original, but this is expected since low noise components most often are more expensive. The factor of noise performance increase is still a lot higher than the factor of the price increase.

The final circuit was never built since the test PCB could confirm that this

final circuit would perform adequately.

5 Discussion

In the beginning there was some trouble with uneven measuring results that would differ from day to day and week to week. It did not differ much, approximately 1-3 %, but it was enough for me to eventually realize that this was probably due to the daily difference in temperature and humidity.

With the makeshift Faraday cage it became better. This was most likely due to the environment in the Faraday cage being more stable especially after a while when the heat from the PCB warmed up the air inside.

This was no major problem to begin with since the difference was minor and the final difference in performance between the circuits was of the or- der of magnitude. The main task was never to get exact measurements of the circuits (but of course the goal was to have as good measurements as possible), but to find a really good and low noise solution to the problem.

After introducing the Faraday cage, the problem was near unnoticeable, and deemed adequate.

The noise of the LT1677 with a gain of one is higher in the figure 4.1 than in figure 4.4. This is because the measurement of the different gains was done early in the process. Later it was found that the values of the resistors in the gain network of the voltage reference OP-amp caused a large increase in noise, probably because to much current was flowing in the feedback loop.

After fixing this the noise went down by a great deal, but there was no time to do new gain comparing measurements. Since the gain comparing only meant to compare the difference in noise between different gains the true value of the noise was not of great importance.

It was found strange that the measurements on the DACs did not show

anywhere near a 35 times difference in noise performance, as specified in the

datasheet. The first thought was that the DACs were specified for the noise

of an internal voltage reference, but none of the DACs features one. From

mail correspondence with Analog devices, the suspicion of something being

wrong with noise specifications in the datasheet was dismissed. This left

the conclusion that the reason for the smaller than expected performance

difference between the DACs is due to the noise being so low that the noise

from the voltage reference becomes the dominating noise.

5.1 Suggested improvements

The measurement equipment is able to measure frequencies down to 0.1 Hz, but when this was tested the measurements was not found to be ac- curate. The main suggested improvement would be to research why the measurements became inaccurate when measuring down to 0.1 Hz with the equipment used. Maybe to use another measurement equipment that could handle measurements down to 0.1 Hz correctly would be the best idea. This would give a more accurate result, especially since most manufactures spec- ify the noise from 0.1 Hz and up.

Another suggestion is to use a more capable CAD software. Eagle was simple and free but had some faults. One of them was the function that should give warnings of non-connected polygon islands. This did not work correctly leaving one ground polygon floating, which caused hours of troubleshooting.

It was also hard to construct good looking and easy to read schematics in Eagle which was a drawback.

If one would like to improve the noise even further, a solution would be to

make use of parallelization. Since noise is added as in equation 2.1 in section

2 and the signal added as the number of devices times the signal due to them

being completely correlated, using for example 4 LTC6655 in parallel would

lower the signal to noise ratio by a factor two. Although this would open

up for extremely low noise this is not cost, space or energy efficient.

6 Conclusions

Noise is not a trivial problem, it is difficult and takes time to measure, the source of the noise is hard to find, it is difficult to know what impact the noise will have on the circuit and it is not easy to remove. However, some steps and precautions can be taken to lower the noise.

The choice of components is extremely important, along with knowing where the sensitive parts of the circuit is. I found that the circuitry around an amplifier is the most sensitive part to external noise. The most sensitive part in my circuit is the Sallen-Key capacitor between the non-inverting input and ground. All noise from this capacitor is amplified, making the use of a ceramic capacitors a poor choice due to the piezoelectric effect. The piezoelectric effect makes the circuit sensitive to the slightest vibrations and increases the overall noise.

This new circuit fulfills the necessary requirements of being:

Ultra low noise

Ten times lower RMS noise over 1-100 Hz (full noise performance can be seen in figure 4.7 and table 4.1).

When looking at all the datasheets of the components used and summarize the specified noises with equation 2.1 in section 2 at a specific frequency, the theoretical result and measured result is very close (for very low frequencies the correlation in the 1/f noise takes over and the calculations is no longer accurate).

Low energy consumption

Approximately the same energy consumption as the control circuit (the new circuits energy consumption is in theory approximately 9% higher).

Cost

The new circuit is more expensive then the old circuit, but within an ex-

pected range since low noise components are generally more expensive than

standard components. The prize for the old circuit is approximately 280

SEK, while the final circuit is approximately 500 SEK when buying the

components to one PCB (the actual price will vary a lot depending on when

and from were the components are purchased. This estimate is made with

prizes from Farnell element14).

Producibility

The new circuit uses fewer components than the old and all of them are SMDs, which means that the PCB can be constructed by pick and place robots. This makes mass production easy.

6.1 Future work

Towards the end of the project it was discovered that there was no need for the DAC to have a span of 0-3.6V but could actually have a span of 0-2.5V which would mean that the LTC6655 could be used as a direct reference to the DAC. This would remove the reference voltage OP-amp and thus lower energy consumption, cost and footprints on the PCB. Reducing the span also means more accuracy as the value of the LSB gets lower.

Another factor to look into would be the use of digital potentiometers in- stead of a DAC. This was only briefly researched because the existing digital potentiometers did not have sufficient temperature or accuracy characteris- tics. Future work may include looking into newer components and different circuits, to increase the temperature and accuracy characteristics, since they generally have good noise characteristics.

Future work could also include the test of using a dual power supply. This

would open up for the use of other components such as dual supplied OP-

amp, giving a larger selection of components to chose from. A more general

idea would be to choose all the components and then adjust the power supply

accordingly.

References

[1] E. Milotti. “1/f noise: a pedagogical review”. In: ArXiv Physics e- prints (Apr. 2002). eprint: physics/0204033. url: http://arxiv.

org/abs/physics/0204033v1.

[2] H. Budzier and G. Gerlach. Thermal Infrared Sensors: Theory, Opti- misation and Practice. Wiley, 2011. isbn: 9780470976753.

[3] Marc de Jong. Sub-Poissonian shot noise. Accessed: 2016-01-27. Aug.

1996. url: https : / / www . lorentz . leidenuniv . nl / beenakkr / mesoscopics/topics/noise/noise.html.

[4] D.W. Ricker. Echo Signal Processing. The Springer International Se- ries in Engineering and Computer Science. Springer US, 2003. isbn:

9781402073953. url: https://books.google.se/books?id=NF2Tmty9nugC.

[5] National Instruments. Understanding FFTs and Windowing. Accessed:

2016-04-12. url: http://www.ni.com/white-paper/4844/en/.

[6] Dr. Michael Belman. Selecting resistors for preamp, amplifier and other high-end audio applications. Accessed: 2016-03-18. Aug. 2010.

url: http://www.eetimes.com/document.asp?doc_id=1278251.

[7] Jan-Willem Pustjens Nebojsa Mrmak Paul van Oorschot. Capacitor Guide. Accessed: 2016-05-10. url: http : / / www . capacitorguide . com.

[8] Ted Jacobson. Complex impedance method for AC circuits. Accessed:

2016-02-16. 2002. url: http://web.mit.edu/2.14/www/Handouts/

PoleZero.pdf.

[9] Texas Instruments. Analysis of the Sallen-key Architechture. Accessed:

2016-02-09. Sept. 2002. url: http://www.ti.com.cn/cn/lit/an/

sloa024b/sloa024b.pdf.

[10] Maxim integrated. A Beginner’s Guide to Filter Topologies. Accessed:

2016-02-10. Feb. 2003. url: https://www.maximintegrated.com/

en/app-notes/index.mvp/id/1762.

[11] Jim Karki. Active Low-pass Filter Design. Accessed: 2016-02-09. Sept.

2002. url: http://www.ti.com/lit/an/sloa049b/sloa049b.pdf.

[12] Stanford Research Systems. Stanford Research Systems SR560. Ac- cessed: 2016-02-19. Aug. 2013. url: http : / / www . thinksrs . com / downloads/PDFs/Manuals/SR560m.pdf.

[13] National Instruments. Benefits of Programming Graphically in NI Lab- VIEW. Accessed: 2016-05-02. June 2013. url: http://www.ni.com/

white-paper/14556/en/.

[14] National Instruments. NI-USB 6361. Accessed: 2016-02-09. url: http:

//sine.ni.com/nips/cds/view/p/lang/sv/nid/209073.

[15] Total Phase. AARDVARK I2C/SPI. Accessed: 2016-02-15. url: http:

//www.totalphase.com/products/aardvark-i2cspi/.

[16] Linear Technology. LTspice. Accessed: 2016-05-19. url: http://www.

linear.com/solutions/search.php?tid[]=9&aid=2125&fid=0&

pid=.

[17] Analog Devices. ADA4528 Datasheet. Accessed: 2016-03-21. url: http:

//www.analog.com/media/en/technical- documentation/data- sheets/ADA4528-1_4528-2.pdf.

[18] Linear Technology. LTC6655 Datasheet. Accessed: 2016-03-21. url:

http://cds.linear.com/docs/en/datasheet/6655fe.pdf.

[19] Analog Devices. AD5066 Datasheet. Accessed: 2016-03-21. url: http:

//www.analog.com/media/en/technical- documentation/data- sheets/AD5066.pdf.

[20] Analog Devices. AD5541 Datasheet. Accessed: 2016-03-21. url: http:

//www.analog.com/media/en/technical- documentation/data- sheets/AD5541_5542.pdf.

[21] Linear Technology. LT1677 Datasheet. Accessed: 2016-03-21. url: http:

//cds.linear.com/docs/en/datasheet/1677fa.pdf.

7 Appendix

BOM and schematics for the final circuit and the test circuit.

Final circuit

Part Value Package MPN OC FARNELL

C1 0.1uF C0603 0805B104K500CT 2496944

C2 1uF C0603 C0603C105K3RACTU 2118128

C4 0.470uF C0603 MCSH18B474K160CT 1856338

C5 10uF C0603 MC0603X106M6R3CT 1759393

C6 0.1uF C0603 0805B104K500CT 2496944

C7 0.470uF C0603 MCSH18B474K160CT 1856338

C9 0.470uF C1206 ECPU1C474MA5 9694293

C10 0.470uF C0603 MCSH18B474K160CT 1856338

C11 0.1uF C0603 0805B104K500CT 2496944

C12 0.1uF C0603 0805B104K500CT 2496944

C13 10uF C0603 MC0603X106M6R3CT 1759393

C14 0.1uF C0603 0805B104K500CT 2496944

C15 0.1uF C0603 0805B104K500CT 2496944

C16 22uF 085CS 1AR TAJB226K016RNJ 197294

C17 22uF 085CS 1AR TAJB226K016RNJ 197294

C23 4u7F C0603 C0603C475K8PACTU 1572625

C24 4u7F C0603 C0603C475K8PACTU 1572625

R2 27k R0603 CPF0603B27KE1 1697395

R3 27k R0603 CPF0603B27KE1 1697395

R4 6K19 R0603 RP73PF1J6K19BTDF 2116762

R5 13K3 R0603 RP73D1J13K3BTDG 1752602

R6 1R R0603 MCHP03W8F1000T5E 1576263

R7 27k R0603 CPF0603B27KE1 1697395

R8 27k R0603 CPF0603B27KE1 1697395

R9 1R R0603 MCHP03W8F1000T5E 1576263

R15 200k R0603 PCF0603R-200KBT1 1160402

R16 470k R0603 CRCW0603470KFKEA 1469814

R17 56k R0603 CRCW060356K2FKEA 1469827

R18 40.5k R0603 MC0063W06031261K 1171026

R19 1K R0603 CRCW06031K00FKEA 1469740

U1 ADA4528-1ARMZ SOP65P490X110-8N ADA4528-1ARMZ 1897114

U2 AD5066ARUZ SOP65P640X120-16N AD5066ARUZ 1827268

U3 ADA4528-1ARMZ SOP65P490X110-8N ADA4528-1ARMZ 1897114

U4 LTC6655CHMS8-2.5PBF SOP65P490X110-8N LTC6655CHMS8-2.5#PBF 2295509

U6 LT3042EMSEPBF SOP50P490X110-10N LT3042EMSE#PBF 2475652

Test circuit

Part Value Package MPN OC FARNELL

C1 0.1uF C0603 0603YC104KAT2A 1327679

C2 0.1uF C0603 0603YC104KAT2A 1327679

C3 0.470uF C0603 C0603X474K4RACTU 1414037

C4 0.470uF C0603 C0603X474K4RACTU 1414037

C5 10uF C0603 0603ZD106MAT2A 1867954

C6 0.1uF C0603 0603YC104KAT2A 1327679

C7 0.470uF C0603 C0603X474K4RACTU 1414037

C8 0.1uF C0603 0603YC104KAT2A 1327679

C9 0.470uF C0603 C0603X474K4RACTU 1414037

C10 0.470uF C0603 C0603X474K4RACTU 1414037

C11 0.1uF C0603 0603YC104KAT2A 1327679

C12 0.1uF C0603 0603YC104KAT2A 1327679

C13 10uF C0603 0603ZD106MAT2A 1867954

C14 0.1uF C0603 0603YC104KAT2A 1327679

C15 0.1uF C0603 0603YC104KAT2A 1327679

C16 0.1uF C0603 0603YC104KAT2A 1327679

C17 0.1uF C0603 0603YC104KAT2A 1327679

C18 0.470uF C0603 C0603X474K4RACTU 1414037

C19 0.470uF C0603 C0603X474K4RACTU 1414037

C20 0.470uF C0603 C0603X474K4RACTU 1414037

C21 10uF C0603 0603ZD106MAT2A 1867954

C22 10uF C0603 0603ZD106MAT2A 1867954

C23 4u7F C0603 0603ZD475KAT2A 1833806

C24 10uF C0603 0603ZD106MAT2A 1867954

C25 4u7F C0603 0603ZD475KAT2A 1833806

C26 0.1uF C0603 0603YC104KAT2A 1327679

C27 0.1uF C0603 0603YC104KAT2A 1327679

R1 5.6k R0603 CRCW06035K60FKEA 1469820

R2 27k R0603 CRCW060327K0FKEA 1652864

R3 27k R0603 CRCW060327K0FKEA 1652864

R4 2.7k R0603 CRCW06032K70FKEA 1469768

R5 5.6k R0603 CRCW06035K60FKEA 1469820

R6 1R R0603 MCWR06W1R00FTL 2447289

R7 27k R0603 CRCW060327K0FKEA 1652864

R8 27k R0603 CRCW060327K0FKEA 1652864

R9 1R R0603 MCWR06W1R00FTL 2447289

R10 12k R0603 CRCW060312K0FKEA 1652834

R11 1.5k R0603 ASC0603-1K5FT5 2078908

R12 5.6k R0603 CRCW06035K60FKEA 1469820

R13 47k R0603 CRCW060347K5FKEA 1469813

R14 1R R0603 MCWR06W1R00FTL 2447289

R15 200k R0603 CRCW0603200KFKEA 1469776

R16 470k R0603 CRCW0603470KFKEA 1469814

R17 56k R0603 CRCW060356K0FKEA 2138473

R18 38k R0603 MC0063W0603138K3 1170945

R19 1k R0603 CRCW06031K00FKEA 1469740

U1 ADA4528-1ARMZ SOP65P490X110-8N ADA4528-1ARMZ 1897114

U2 AD5066ARUZ SOP65P640X120-16N AD5066ARUZ 1827268

U3 ADA4528-1ARMZ SOP65P490X110-8N ADA4528-1ARMZ 1897114

U4 LTC6655CHMS8-2.5PBF SOP65P490X110-8N LTC6655CHMS8-2.5#PBF 2295509

U5 LT1677CS8PBF SOIC127P600X175-8N LT1677CS8#PBF 1330687

U6 LT3042EMSEPBF SOP50P490X110-11N LT3042EMSE#PBF 2475652

U7 AD5541AARMZ SOP50P490X110-10N AD5541AARMZ 2096192