Noise Analysis of Trans-impedance Amplifier (TIA) in Variety Op Amp for use in Visible Light Communication (VLC) System

Received Jun 19, 2017 Revised Oct 20, 2017 Accepted Dec 4, 2017 VLC is a complex system with lots of challenges in its implementation. One of its problems is noise that originated from internal and external sources (sunlight, artificial light, etc). Internal noise is originated from active components of analog front-end (AFE) circuit, which will be discussed in this paper, especially on the trans-impedance amplifier (TIA) domain. The noise characteristics of AFE circuit in VLC system has been analyzed using the variety of available commercial Op Amp and different types of the photodiode (Si, Si-PIN, Si APD). The approach of this research is based on analytical calculus and simulation using MATLAB®. The results of this research show that the main factor that affecting the noise is en, the feedback resistor (Rf), and junction capacitor in the photodiode (Cj). Besides that, the design concept of multi channel TIA (8 channel) using IC Op Amp, with consideration of pin number of each Op Amp, supply needs, the initial value of Rf, converter to 8-DIP and feedback capacitor (Cf) channel, also discussed in this paper. Keyword:


METHODS
The steps of this noise simulation experiments is shown in Figure 1, where there are 12 steps, starting from modelling of the overall VLC system, photodiode selection, considering amplifier IC types from several IC producers, selecting the Op Amps, mapping the Op Amps specification into the tables, calculate R f , calculate C f , calculate noise on TIA circuit, plotting the parameters, then buying and stocking the components, develop the DIP-8 IC converter and the last is developing the multichannel TIA kit. We use the eight Op Amps from the different manufacturer to observe the application potential of each IC with different characteristics.
In line with the experimental procedure that shown in Figure 1, the first step of this experiment will be explained in section 2.1., the second step will be explained in section 2.2., then the third, fourth and fifth step will be explained in section 2.3., The most important step of this experiment, the sixth, seventh and eighth steps, where the mathematical equation is done, will be explained in section 2.4., The results and analysis then will be explained in section 3.   Figure 2 shows general OWC link for intensity modulation based VLC which cover system block, interferences signal & external noises, internal noises and challenges in VLC system (e.g. Fading and shadowing). We adopted Figure 2 from Karunatilaka, et al [6] link structure and then mixed with noise models from K. Sindhubala, et al [7], major limitation model in VLC system from N. Noshad, et al [8] and noise models from Orozco [9]. On the channel parts, it is shown that as long as light propagating on the free space, the signal will be affected by interferences, such as from intercell-VLC interference [10], other noninformation light sources [11], shadowing effect caused by passive blocking object (e.g. indoor appliances or active blocking object), by people moving or human blocking [12], [13], and fading, which caused by uneven distribution of light and distance from the transmitter to receiver.
The detailed explanation of Figure 2 is not included in this paper. The focus of this paper is on AFE domain, especially the TIA domain. The photodetector receives information light from LED and generates I rec current, which then converted into V rec voltage by the amplifier. The received signal on this photodiode contains noise, i.e. the external source noise from the environment or ambient light (sunlight, incandescent, fluorescent, flashlight and generic LED lamp). Then, on the AFE domain, noise is commonly seen in active component or amplifiers such as shot noise/Schottky, flicker noise (1/f), thermal noise/ Johnson-Nyquist noise (among other names) and nature quantizing fluctuation. Shot noise is generated by the statistical fluctuations of currents in the active components. Thermal noise is due to its equivalent resistance and capacitance. On the photodetector side, noise also generated (e.g. Dark current and optical excess noise) [7]. These parameters should be added to the calculation to find the ideal received signal (V sig" ) which then would be processed on the demodulator (digital domain). Typically, noise consists of voltages and currents which flow in the active (e.g. Op Amp) and passive circuits (e.g. R, L, C) which can be expressed as noise density (E n and I n ). On the TIA domain, based on [9], there are 3 noises i.e. R f noise (N Rf ), current noise (N current ), and voltage noise (N voltage ). We must also include noise from the extra capacitance (C c ) if the circuit is implemented on the project board or using selector switch and I/O terminal. Figure 3(a) adapted from [14] shows the TIA circuit with the photovoltaic topology which is used in this experiment. Figure 3(b) is an equivalent model of TIA (R sh on the photodiode is ignored). In this paper, the noise in TIA circuit based on the root mean square (RMS) of these three noises is analyzed.

Selected Photodiode
VLC is one of the OWC which uses visible wavelength with the electromagnetic spectrum of 380nm -780 nm [15][16]. To receive a signal from that spectrum, we need to use photodetectors, such as phototransistor or photodiode. The Photodiode is more recommended because it has better linearity, dynamic range, and stability. It also has a wide operation wavelength, 200 nm -2000 nm which also includes visible light wavelength. Although, the phototransistor has advantages on its low-cost price [17]. For experimental needs, we use three types of the photodiode. All of those are manufactured by HAMAMATSU, in which the types are Si PIN (320 nm -1100 nm), Si APD (200 nm to 1000 nm) and Si (380 nm to 780 nm).
Exploration of many photodiode types is also an open research topic in VLC [11], in which we also have already discussed this topic in another paper [18], [19]. But, we use photodiode only to find the intrinsic capacitance variable (C in ) to analyze the noise characteristic on TIA domain. So several factors, such as area, 163 rise time, cut-off frequency, dark current and short circuit current are not included as the main consideration, even though on the real implementation, those factors must be considered. The first step of this experiment is by choosing three different types of the photodiode, especially with significant differences on pF parameter, i.e. 3 pF, 30 pF, and 150 pF. Detail of photodiode specification is listed in Table 1.

Selected Op Amp
According to the introductory paragraph, there are several Op Amps criteria which need to be considered on TIA design, those are: a) High Bandwidth, b) Low Noise, c) Low power consumption, d) has good stability and sensitivity, e) large gain, then the last one is f) that components are commercially available in the markets.
To obtain those features, general Op Amp can"t be included in Op Amp selection for TIA. Op Amp IC as the main variable is randomly chosen from several IC manufacturer which produce "amplifier and linear circuits". Those manufacturers are Maxim integrated circuits®, Analog Devices®, Texas Instruments®, Linear Technologies®, Microchips®, ROHM®, Intersil®, and Fairchild or ON semiconductor® where each manufacturer produces Op Amp with suitable features to develop the TIA circuit. Specification detail of each manufacturer is shown in Table 1 which contains the specification of MAX9637, AD8014, OPA380, LTC6268C, MCP651, BA2107G, ISL55001 and FAN4852 from its datasheet.
There are 14 variables that needs to be considered, those are IC manufacturer, IC series, packaging type, quantity of Op Amp in one chip (channel), intrinsic capacitances of IC (C in ) in pF, input current noise of IC (E n ) in nV/ , input N voltage of IC (I n ) in fA/ (pA units must be converted to Femto ampere for ISL55001 and AD8014), maximum positive supply voltage (V+) to minimum negative supply voltage (V-), output voltage swing capabilities (V out ), gain of IC (A VOL ) in dB unit or V/V, gain bandwidth product of IC (GBW) in MHZ, slewrate IC in V/µs unit, power consumption in mWatt and input bias current (I B ). Cost factor ($) is not considered in this experiment.
From Table 1, it can be shown that not every parameter is available on the datasheet, e.g. BA2107G, in which C in and I n is not available. So, for simulation needs, C in is assumed to be 1 pF, and I n is assumed to be 10 fA/ . The assumption is based on the comparison with OPA380. The approximation of I n and C in value can be done by comparing it with MCP651 specification with e n of around 7nV/ . But, for heterogeneous of simulation data, we choose to approximate those values from the characteristic similarity with C in and I n from OPA380.
The most important variables for this noise characteristic simulation is E n , I n , and C in from each Op Amp IC. Bandwidth target of this OWC project is ~1 MHz, therefore capabilities and range of operation of Op Amp has to be larger than the target bandwidth. In this case, MAX9637 with gain bandwidth product (GBW) of 1.5 MHz is acceptable. According to [20], that besides I n , E n and C in which is the input noise, there are other parameters that need to be considered in Op Amp selection, those are: a) large open loop gain, so that sensitivity of TIA is not affected by external capacitor that connected to Op Amp (minimal >50dB), b) low input bias current, to minimize the dc output error and precise setting of R f (in pico-Ampere scale), c) bandwidth and slewrate, with large bandwidth amplifier stability can be maintained with ideal capabilities of Op Amp for high-rate OWC such as VLC is >100 V/µs. Then, the limitation of Op Amp that is chosen for this simulation experiment are a) there are 5 Op Amps with <100 V/µs; d) Gain and A VOL of AD8014 is not available on the datasheet and e) power consumption characteristic data is not available. Therefore, input noise is used as main parameters for the simulation and implementation. Table 2 shows the specification of Selected Op Amp  From those eight chosen IC, there are three IC which has 2 channel, those are FAN4852, BA2107G, and MAX9637. The others of the IC only have 1 channel. So that the circuit can be implemented as a plug and play to the TIA board, we need IC pin converter to DIP-8, e.g. SOIC-8 to DIP-8, MSOP-8 to DIP-8, etc. Based on international reference guide, IC packaging of 8-pin Op Amp, the non-inverting pin is at number 2 and the inverting is at number 3. Whereas, the output is at number 6 for 1 channel Op Amp.
From the supply consideration, IC from Maxim and ON semiconductor only needs one input supply (single supply), those are MAX9637 (+2.1 V DC to +5 V DC ) and FAN4852 (+3.3 V DC to +5 V DC ). The others of the IC can be used for double supply. The planning of TIA board based on the specification of the above 8 chosen IC, needs the single/double power supply and the number of the channel on Op Amp IC. The board or kit that wants to be developed is "8-channels input TIA", which concept is shown in Figure 4. We use a variable resistor with value of 100 kΩ so that the output voltage can be tuned as desired.

Experiment Procedure
TIA or photodiode amplifier consists of single/array photodiode, IC Op Amp, R f , and C f . The key elements of this design are the photodiode & Op Amp selection and then R f & C f calculation. The working principles of TIA is converting the light signal received by photodiode into current, then that weak current is flowing through the resistor, converted into voltage and amplified by the Op Amp. The general equation of TIA is V out = I pd * R f . The resistor serves as a gain of the amplifier because I pd is constant and limitation of the photodiode to generates current, depended on the intensity of the light. The value of the resistor can be determined corresponding to the maximum current of photodiode and desired V out (Equation (1)). The larger the value of the resistor, the larger the value of V out [21], but gain determination has to be done carefully on TIA design for the OWC. Because it can affect the bandwidth to be narrow and the signals can be clipped. The detailed description of bandwidth is beyond the scope of this paper.
The capacitors also have important role in reducing the noise, optimizing the TIA response, determines the stability of TIA circuit, and also in reducing the overshoot. The larger the C f , the slower the response of the TIA and the smaller the C f , so that the amplifier is oscillating, even though it increases the bandwidth. The value of C f forms a pole in the frequency response with 45 o of phase margin, is given by Equation (2), where C j is photodiode"s shunt capacitances in Farad and C in is Op Amp input capacitances in Farad. Therefore the maximum value of C f can be determined from the desired bandwidth (in this case we use 1 MHz frequency as gain bandwidth product (GBW). fGBW * R * 2π Then the peak noise can be determined based on Equation (3). With assumptions in which the first zero and pole of the noise density output less a decade lower than the second pole and which the output noise is equal to the plateau noise (N 2 = plateau noise), where e n is the voltage noise density obtained from the datasheet.
The calculation of N voltage is based on Equation (4), where N 2 is obtained from Equation (3) and f p2 are obtained from Equation (5). Then the N voltage can be rewritten as Equation (6), The total sum of C j , C in , and C f is C tot or C t in Farad.
The Op Amp"s current noise will appear on the output after going through the R f , given by Equation (7), where I n is noise input density obtained from each Op Amp datasheet. ENBW, which is equivalent of noise bandwidth can be obtained using Equation (8). The calculation of TIA"s cut-off frequency (f -3db ) is based on Equation (9).
The noise in R f of TIA is given by Equation (10). Where k is the Boltzmann (1.38 × 10 −23 ) constant and T is the temperature in Kelvin (283 K).
The three noise sources in the mathematically calculation are independent. Gaussian means that the N total is the root-sum-square (RSS), which given by Equation (11). The specific circuit of Low Pass Filter (LPF) on the TIA output can greatly reduce the N total if f p is much higher than the signal bandwidth by adding singlepole RC filters.

RESULTS AND ANALYSIS 3.1. Impact of the Higher R f Selection
In this experiment, we use one type of photodiode as the parameters, that is silicon (Si) S9219-01 with characteristics Cj = 150 pF and one type of Op Amp IC, that is OPA380 "Precision, High-Speed Transimpedance Amplifier" with characteristics: C in = 1.1 pF, E n = 5.8 nV/ and I n = 10 fA/ . Because of the limitation of the space for the figure of the results, we use only four resistors with different values, those are 0.5 kΩ, 1 kΩ, 1.5 kΩ and 2 kΩ. This simulation is using the calculation based on Equations (6), (7), (10) and (11). According to Figure 5, it can be shown that the value of R f is linear against the noise. The larger the value of Rf, the larger the value of N voltage , N current , N Rf which affects the N total . This is consistent with the experimental results of [22] which prove the relation between gain and RMS noise. From this experiment, it can be concluded that R f is one of the main contributors to output noise.  Figure 5. Relationship between noise (voltage, current, R f ) and noise total at different R f (0.5 kΩ, 1 kΩ, 1.5 kΩ and 2 kΩ) when using OPA380 Figure 6, Figure 7 and Figure 8 shows the characteristics of the eight investigated Op Amp. The Y-axis is scaled from -5 to 75, because of the limitation of the space and it has a very wide gap (extremely low and extremely high) where the detail value can be shown in the table below the figure. The X-axis shows the observed variables and the corresponding units. Variables e n , i n and C in are obtained from the datasheet of each Op Amp as fixed variable, whereas C f , N Rf , N current , N voltage , and N total is based on the calculation of the section 2 point D. The R f and GBW are independent variables.

Impact of the Difference Photodiodes
Because on section 3 point A the simulation has been done for R f = 2 kΩ so that the simulation results are not repeated (heterogoneous of data), in this pont we use R f = 2.2 kΩ. The GBW is configured the same, 1 MHz. All of the noise is on micro (µ) units, except for N current which is in nano (n) unit due to very low, and it is hard to see the gap of the low noise scale. For example, the differences between N current which are insignificant using the micro scale, i.e. LTC6268C = 0.031 with BA2107G = 0.05685 with the maximum scale of 75. This will be easier to be shown if using nano scale, i.e. 31 and 56.
By using three types of the photodiode with the variated gap of C j , we obtained interesting results. From the three figures, can be concluded that: a) the larger the e n , the larger the N voltage and the N total . It can be shown on MAX9637, even though it is really low. b) The smaller the i n the N current tends to be low, it can be shown from the significant differences between MAX9637 with OPA8014 and ISL55001. c) The value of C j affecting the C f significantly, that is the larger the C j , the larger the C f , and so does the opposite.  Figure 9, Figure 10 and Figure 11 show the graphics of N total in µVRMS against frequency. The purpose of this experiment is to find the N total behavior by variating the frequency, from 1 MHz (the minimal bandwidth target) to 10 MHz, with interval of 2 MHz and the value of R f is constant 1 kΩ. Because of the space limitation for the figure of experiment results, in this point we only show the N total , so N voltage , N current , and N Rf are not simulated. From Figure 9, Figure 10 and Figure 11, it can be observed the N total of each IC Op Amp is linear against the frequency range. The higher the frequency, the larger the N total . The difference in the value of C j of photodiode also affecting the noise, the larger the C j , the larger the noise.

Frequency and Cj as a Function of Total Noise (N total )
The graph shown the extreme noise and the stable noise. MAX9637 has the larger noise deviation and steep curve, it is because of its e n factor, which is larger than the other Op Amps, around 38 nV/ . So it affecting in the large N total , even though it has the lowest i n , around 0.9 fA/ . Then, AD8014 has the largest i n, around 5000 fA/ , so it has large noise even though the curve is slope, different with MAX9637. After that ISL5501 has an extreme characteristic with significant differences in the i n (1500 fA/ ) with C in (1pF) and e n (12 nV/ ). So, the N total is tends to be large, but with different graph form with AD8014 which has the combination of low C in for large noise. LTC6268C has low N total , it is because it has lower e n, i n and C in than the others Op Amp. This experiment is only to shows the effect of noise, not for comparing performances of each product of IC manufacturer.

CONCLUSION
In the last of several decades, the wireless communication has been adopted in many application [23][24][25][26]. VLC is one of the wireless communication optics-based that also popular to be implemented in our life, such as for indoor positioning system [27], audio transmission system [28] and etc. This technology can cover RF communication that can"t exist in the closed area. In this paper, we focus on noise analysis in VLC system.
The noise characteristics of Op Amp for VLC applications based on analytical calculus has been described. According to the experiments, it can be shown that the main contributor to the output noise is the "voltage noise density" (e n ) which comes from the Op Amp itself and from the external components, i.e. R f . This experiment"s results are consistents with results of [29], [30]. Besides that, the Cj, which is an internal characteristic of the commercial photodiode, also affecting the noise. The larger the Cj, e n , and Rf, the larger the N total is. The concept of the fabrication method of multi-TIA with total Op Amp of 8 also has been explained, with consideration in the total amount of pin IC, amount of Op Amp in one IC, packaging type of IC and the amount of power supply. After this, we will assemble the real system based on discrete components, so we will be able to provide comparisons of the real implementation with the simulation results, which also be interesting as the topic of the next publication.