High Resolution 24-bit Daq Card Fits Into Standard Sdroms Slot

White Paper: High resolution 24-bit DAQ card fits into the standard CDROM’s slot. G. Rozite1 and A.Murashov1 1 Institute of Physical Energetics 21 Aizkraukles Str., Riga LV-1006, LATVIA [email protected]

Abstract Most commercially available data acquisition cards have either low resolution (<16-bits) or are made on audio codec (ADC) chips, with a very poor DC accuracy (although a 24-bit resolution), and are not applicable in places requiring high dynamic range and low noises, especially in the 0.1-10 Hz region. Here a new concept of high resolution auto-ranging 10V – 10mV PC card for demanding scientists is presented. The card supports plugging remote amplifiers/conditioners as well as remote power and biasing. Such unique features of the card make it applicable for a wide range of scientific tasks, e.g. in laser interferometers, lockin amplifiers, for spectrum measurements, phase demodulation, statistical deconvolution, etc. The architecture proposed by the authors is flexible, because most of the tasks are done in DSP, while the analog part is realized on the 4th order ADS1254 super-high resolution delta-sigma ADC, which became available only in 2008. The authors discuss the functionality of the proposed card and consider the implementation details. The concept has been created as the result of multiple cross-correlated user requirements. The proposed hardware concept is royalty-free and can be employed under the GNU public license; however the firmware should remain commercial to promote development effort. Key words: 24-bit DAQ USB card, USB seismometer, USB Laser vibrometer, Electrochemical impedance spectroscopy.

Introduction The only known supplier of 24-bit DC measurement PC cards is the National Instruments with their http://www.ni.com/dataacquisition product range. If your product is not there, then it probably does not exist at all. In the past, our laboratory bought two Keithley 2000 digital multimeters with the aim to employ them as channel digitizers. However, the SCPI language, (www.scpiconsortium.org ) in which most digital instruments (including Keithley 2000) communicate to the host, can send data in retriggerable packets. Such serial communication (RS232) takes ~160 ms to start new measurement series and does not allow continuous stream. If we want to increase speed and acquire sample by sample, then we lose resolution. Another approach we have tried to employ was 24-bit sound cards (E-MU 1212M series). With a minor tuning it was possible to provide pure DC measurements preserving full AK5394 chip speed. However a new problem arose - thermal fluctuations, which reduced the ADC chip performance, so only a massive heat sink on AK5394 chip allowed us to obtain ~25uV noise plus offset from our E-MU 1212M series sound card (or 5 digits resolution). Interesting enough, the chip temperature was affected also by the output code warming chip up! Besides the heat sink, also large linear regulators were employed to supply low-noise highstability ±12V and 5V power to E-MU 1212M. The results of experiments allow us to conclude that so far the scientific tasks [1] have not been provided with high resolution measurement cards. The proposed PC cards concept based on new 4th order ADS1254 super-high resolution delta-sigma ADC aims to fill the gap in this growing market segment.

Functionality The proposed concept of USB card is illustrated by Figs 1-4. Block diagram in Fig.1 gives a brief overview of the analog in/outs, DSP processor and front panel controls. Following the design of conventional PC sound cards, also here should be bi-directional audio streams – up-stream to host and down/control stream to the USB card. That can easily be achieved by AT91SAM9XE DSP processor dual USB ports seating on a motherboard’s 10 pin USB header (like a PC internal 3.5”card reader). With the basic firmware version an AT91SAM9XE DSP processor serves as a stream router, and the PC card works much like a conventional sound card with two 24-bit inputs, two 16-bit voltage outputs, and one current output. In more sophisticated applications DSP can be used as the optical phase demodulator, thus forming a high-resolution laser seismometer, or a nm-range displacement sensor.

Fig.1 Block diagram of USB card The AT91SAM9XE was chosen because of its cheep development tools, high performance and good Cfriendly programming, allowing the use of embedded Linux as real-time OS. The 16-bit output DACs shown in the figure form a programmable ±10V voltage source (Out A), and a 3A current source (Out B), which can be employed as a precision battery charger in the electrochemical impedance spectroscopy (EIS) applications. The dual A Outs are tied in a bridged configuration and provide differential outputs suitable for driving a wide range (with only the load impedance limited to >4ohm) of piezo- and magnetic actuators. They are low-noise (<1uV) close to ideal voltage sources suitable also for reference biases. Galvanic isolation from the noisy PC chassis ground is realized through the transformer coupled power supply and USB isolation, which is rather tricky. Another way would be to isolate control wires in DAC/ADC side. The ADC sampling frequency can be PLL-locked to the external clock input by CS2000 (Cirrus Logic). The

presence of a laser source requires two additional connectors: DIO and LASER driver. The DIO connector forms a serial bridge, while the second connector supplies high-stability power to external laser controller. The overall line regulation for +5V and dual ±7.5V is ~120 dB, which is a very good result.

Fig.2 24-bit ADC input and bias pedestal. The ADS1254 converter allows only unipolar 0…5V input (which is enough for scientists working with photodiode signals). With 6 dB attenuators it is possible extend even further the input voltage range – up to 10V (single input configuration), and ±20V common range (in differential configuration). We found that the ADS1254 reference input introduces a very high temperature coefficient of 35uV/C. To overcome this limitation, much attention was devoted to the temperature compensation. The Pi attenuator on top of 4th order ADC forms a TC compensation voltage – 50 uV which trough the 100R NCT is feed into non-inverting input of U10, and finally zero 35uV temperature offset. Also, thermal isolation around the reference U4 and ADCs and DACs (dashed line noted in Fig.1 as Isothermal Block) would increase the overall stability. To reduce thermal fluctuations, all outputs - differential amplifier U9, U1:A BIAS DAC and voltage reference U4 were buffered. Besides, DIP packages would increase the temperature stability, because of better heat dissipation. The biasing circuit is formed by BIAS DAC8831 and two unity gain U1:A, U1:C buffers. The DAC8831EP were chosen owing to their good temperature coefficients. The BIAS DAC8831EP forms the 5V output which is directly fed into the U2 ADC CH1- input. The next 5V BIAS output is attenuated by the order of 10 and recursively fed back to the ground reference, forming a 65mV artificial ground offset. Such arrangement allows coarse and fine bias adjustments, done manually from the front panel, or automatically after pressing ZERO button. The ZERO-ing routine is interesting and needs a deeper analysis. The sampled data from DSP are averaged in a memory accumulator - moving average estimate (see the right most corner in Fig.2). When cashed samples are enough to satisfactorily approximate the mean value; the estimate data are enveloped by hamming windows, and then digitally compared with the cashed samples in estimation array. Only less deviated samples that have passed through the comparator are used for bias compensation. Due to the

ADS1254 ADC chip traits, the cards 20V input range is unipolar, though can be shifted by using internal MUX, however this is impractical. Instead, in the proposed concept we assume that the U9 INA115 output is always positive, whereas the negative voltage will be considered as an over-range error. Internal MUX implements a time gate, synchronized to EXT.CLK. An ADS1254-based design can be considered as a DC version, while the ADS1281 providing a true ±2.5V “audio range” input – as an AC version, especially for electromiogram (EMG) or cardiogram (ECG) applications (although it is out of our interests). The U9 1:1000 Programmable Gain Amplifier makes the input range tuning very convenient. When there is a test signal on the input, pressing the RANGE button will start the auto range routine. The measuring device will automatically set the best range for the input signal. Once the range is set, the ADC will perform the conversation of even overdriving inputs. To avoid such overload situations, an internal buzzer and the front panel LEDs will tell the user that the RANGE needs attention.

Fig.3 DAC programmable analog OUT drivers. All the three outputs in Fig.3 are DAC-programmable, and can be employed to generate DDS waveforms. Outputs A+ and A- are a true wide band up to 10 kHz, while the bandwidth of current source B is limited to <10 Hz due to the SEPIC converter large output capacitors. The current source was designed as a battery charger, giving possibility to charge also 24V batteries. The mutually coupled inductors L17:A L17:B can achieve zero ripple condition only at given duty cycle, but 0.25…0.95 duty range in our case would produce high ripple. Therefore SEPIC converter was used together with so-called harmonics rectifier resonating at 307.2 kHz (multiple of ADC sampling frequency), and improving rectifiers efficiency by ~1.25. From rectifiers OUT_B small fraction of HF harmonics returns to ground via node D9, D11, R7, and should be sensed by U2:A feedback amplifier. It can be seen, that current sense resistors R46, R7 and weight resistors R47, R8 are equal, producing total current, which steers U6 UCC38C42 controller and is feed to ADC input (CH_CURRENT). Programming of duty cycle was realized through DAC U2 and buffer Q17. The maximum duty cycle 0.95 corresponds to 0,05 V at the CS pin, while minimum 0.25 - 0,75 V at the same pin. The Vref pin 8 of U6 creates ~40uV noise, and therefore has extensively filtered by a 47mF super capacitor. The voltage divider R2, R3, RV4 senses the charge level,

whereas for better flexibility a 5V CHARGE_V was duplicated also in the OUT B connector, allowing voltage sensing from the multi-element battery midpoint. It should be remembered that the SENSE input is not overvoltage-protected. Outputs A are formed from two identical AB-class amplifiers with a gain of 2. The analog SW2 (SINGLE) toggles between the bridged and the single ended load, whereas SW3A and SW3B (BIAS_ON) adds Bessel low-pass filtering capability (a 10 s filter) needed for remote sensor_BIAS. Two precision RV2, RV5 potentiometers trim the amplifier gains to the desired value. The cermet trimming resistor RV1 has effect only at the open feedback loop, otherwise the loop gain completely removes any offsets, so such a resistor makes no sense. The output bias, which is typically 15-20 mA, through Q9, Q10 transistors is temperature-compensated via the network including RT1. In the end, a question naturally arises: is it reasonable to use such complicated output drivers if a wide range of audio amplifiers are available – e.g. TDA2040, TDA7490L? Yes these can be used also at the expense of reduced phase margin and linearity.

Fig.4 Transformer isolating power supply. The super-resolution [1] card will not also work with noisy 12V and 5V PC power. A residential 12V ripple varies around 100mV. The motherboard and peripherals of different load conditions increase the voltage uncertainty, whereas HDD adds wide-band noises. The only escape is to keep an on-board power converter (see Fig.4), of approximately 36W power. The development of this power supply was aimed to achieve the best possible line regulation and the minimum ripple noise. It is known that all linear regulators (like 78XX, TL431, LM317) use bandgap references [1], so their outputs are very noisy (~30uV) and has high temperature drift. Besides, these regulators have also very high temperature coefficients. In our design we tried to avoid them as much as possible. Heart of primary side is U3, the Motorola MC34067 newest HBridge controller. The quasi-resonant mode employed here essentially realizes ZVS below the resonance frequency (800 kHz). Note that during normal operation a transformer presents a capacitive load, and the frequency of a higher VCO voltage corresponds to a higher power output. Special attention was devoted to attenuate the most unwanted component – a 100Hz ripple coming from the ATX power supply. By C12, L4 elements shown in Fig.5 the ripple is suppressed by 40dB. Transformer TR2 is wound on an ETD29 core, but, since there is not enough space to wind all required supplies, the digital supplies are made from an intermediate 12V output. Since the switching frequency is very high, the windings should be executed with

litz wire or with parallel strands because of skin effect. Increasing winding density by using copper foils (e.g. www.ppi.ie ) produces more space, but acceptable results can be gotten from 4 copper strips at the least. The 15V and 7.5V cross-coupled filter inductors L5 and L8 add the voltage tracking capability to the linear supply and form close-to-ideal output voltage. The transformer-coupled feedback generator UC2901D is far better than the TL431 opto-coupler counterpart because of its higher bandwidth and lower noise. The linear regulator U4 adds better stability against overload conditions, keeping supply voltage of U5 at a constant level. Secondary side diodes are damped by ferrite beads L6, L7, L10, L11, L14, L15, thus significantly reducing EMI noise.

Elements of design and front panel controls The most advantageous card location is the front panel CDROM slots (a draft is shown on Fig. 5). There is much more room, and, which is most important, there are less EMI noises than in other PC chassis parts. The most common controls are rotating encoder knobs and two resin press buttons used for the auto ranging and automatic offset zeroing.

Fig.5. Front panel outlook (150% scaled). Two LEMO 7P REDEL sockets for remote amplifier inputs A and B allow several custommanufactured amplifiers/conditioners to be plugged in, making the proposed USB card really universal. The remote pin-out has not been yet standardized, however we can expect here that the cable jumper (see PLUGIN in Fig.2) will shorten the negative input to ground if the differential mode is not used. Remote power, bias and one GPIO pin should be enough even for most dedicated applications. The best possible resolution can be achieved by using precise biasing circuit (Fig.2. BIAS DAC, U1:A, Q3, U1:C, SW4) driven from incremental encoders (e.g. www.bourns.com/pdfs/em14.pdf). As one can expect, non-pressed knob rotation will allow fine adjustment, and, while pressed, it will drive the whole 5V bias range in 1 or 2 turns. Every front knob motion should be monitored by an INT event. A rich set of AT91SAM9XE256 GPIO pins does such job very well. It can detect touching of 2 x 3 = 6 encoder lines, 3 encoder press switches and 2 resin buttons. As the encoder reads input through a micro-controller, it can be used for multiple controls. The third encoder has some DSP parameter functions, which will depend on the installed firmware. firmware version Basic spectrum Heterodyne interferometer 2-ch vibrometer /nanosensor

3rd encoder function/ pressed function sensor_BIAS Fine/Rough Integration time, ms /LPF Hz Integration time, ms /LPF Hz

Table.1 DSP parameter functions according to the firmware versions.

green LEDs function under/overrange under/overrange phase state 90/180 deg.

The white coaxial input EXT. CLK. input in Fig. 5 is either ADC time gate used to sample A or B inputs during predetermined time interval, or a PLL reference clock used to generate DDS PZT waveform. The Deflation mirror coaxial audio socket is differential amplifier INA115 output (used in the laser seismometer, see Fig.7). The laser controller is powered from a high-stability low-noise 5V and ±7.5V power sources expanded to the front panel through the 4-pin header. The SYNC RS232 connector makes a fully duplex serial interface with the external laser driver-controller, and the additional SYNC RS232 DIO pins (visible also in Fig.1) can be used to sense or drive the TTL inputs.

Firmware Firmware is known to the most expensive part of every project, therefore its development should rely upon on a commercial basis (e.g. involving a third party’s developers). Real time embedded OS and support library (presenting hardware) are the minimum, that developers could get starting their own scientific application. From the host side its most demanding component is a low-latency ASIO USB driver, which probably has already been written off. ASIO framework is able to handle 8_IN/8_OUT data streams in a parallel manner, which is more than necessary. It has delay time 2 -100 ms (latency), and this time is always constant (for ASIO features see [2]). The data in an ASIO stream are supplied with the so-called time code. Essentially, it is a local session time from the beginning of measurements till the end. Perhaps, a good idea could be supply an ASIO add-in for MathLab. Next, we will discuss how the proposed card can solve scientific tasks by using application specific signal conditions/preamplifiers and firmware. Basic spectrum The basic spectrum is characteristic of OLED and solar materials and generally relates to photodiode signals [3]. The fiber pick-up units and photodiode amplifiers are high-impedance I-V converters powered from a dual 7.5V remote power supply. A BIAS sensor (see Fig.2.) allows measuring I-V dependences directly from the card thus avoiding external wiring. Due to extremely high requirements to photodiode BIAS stability and noise, amplifier A+ (see Fig.3.) was low-pass filtered by a 10 s Bessel filter by setting BIAS_ON.

Fig.6 External Quantum Efficiency measurements for characterization of solar materials. Fig. 6 shows that a collimated light beam coming from the high-intensity discharge lamp through the chopper modulator is fed into the monochromator input slit (typically 0.2 ..2mm wide), which is used as a

feedback actuator keeping the light intensity at the monochromator’s output at a constant level. The output beam from the monochromator through the light guide is applied to the measured PV sample located in the vacuum chamber. A small fraction of the output light is applied also to the light level photo diode used to calculate the light feedback signal and the external quantum efficiency (EQE) described also in [4]. A very important in this setup is the photo-interrupter signal driving lock-in synchronous demodulation realized in a PC processing program (it could also be done in DSP, but due to the availability of existing Windows programs and dedicated algorithms, this part is omitted). Essentially, after both channel demodulations, the reference signal will be subtracted from the light signal giving the pure photo-EDS proportional to the EQE of the PV sample. Synchronism here must be held to a very high degree. To achieve this, a dark/light TTL signal is directed into a separate audio stream with the applied time code attributes! A question naturally arises: is it reasonable to stream TTL signals while the USB events can perform the work? No USB event would arrive to host PC 5-100 ms before the actual sampled data because of the ASIO driver’s latency; but it would be always correct for a streamed version, because it would be repackaged in a synchronous manner (see ASIO features in [9]). Another useful feature of this setup is the ability to take V-I dependences of a PV sample (see also [3]) by using a sensor_ BIAS (Fig.2, 3). The DDS Waveform generation feature is not required in this setup, however it can be employed to drive coils, generate bias voltage, etc. Heterodyne laser seismometer. In Fig.7, the interference pattern (e.g. the pattern from a two-arm Michelson interferometer) is focused onto the mirror of the electric-mechanical deflector. The coming back pattern is displayed on the transparent-reflecting grating (lenses are omitted). The fringes of this grating are parallel to those of the initial interference pattern (e.g. 8 kHz), with their periods coinciding.

Fig.7. Laser seismometer. The servo-loop shown in Fig.8 is used to compensate the slightest shifts of the interference pattern by deflection along the grating, thus forming an optical frequency converter – a heterodyne. The servo-loop output current is proportional to the pendulum displacement. Thus the feedback loop of seismometer is locked [4-6]. In the differential buffer amplifier INA115 the photodiode PDA- and PDA+ signals are subtracted and forwarded to the deflation coil, which is driven by very small current – typically ~200uA. The deflation signal is proportional to the measured acceleration, which further is fed into the integrator (through ADC) and PID controller. This part of feedback compensates the gravitational force applied to pendulum, thus keeping it in a constant (levitating) position. Probably, this is the most time-critical DSP task, because the overall latency must be kept within 1ms, which corresponds to a 1kH bandwidth. The laser-synchronized

PZT waveform generation is not a very time-consuming task, and can easily be run in background. The integration time constant is adjusted by the first from left rotary knob. DDS generated PZT waveform can be synchronized to external clock, e.g. 5 kHz.

Fig.8 Feedback system of the seismometer The force balancing coil driver mentioned above is shown on Fig.8. Here we can see that the velocity signal is made up by integrating the acceleration signal. Further, the velocity signal is fed into Proportional Integral Derivative (PID) controller, ensuring that the force applied to the force balancing coil will not upset the levitating equilibrium [6]. 2-channel Doppler vibrometer - displacement sensor Another application of card is a two-channel Doppler vibrometer. Laser beam is divided into a reference beam and a signal beam. The signal beam is directed onto a vibrating test structure, and the backreflected light is recombined with the internal reference beam. When the test structure moves, the frequency of the signal beam is shifted, resulting in the intensity modulation of the recombined beam due to interference between the reference and the signal beams [7].

Fig.9 2-channel Doppler vibrometer

One complete cycle of the intensity modulation by a 2-channel Doppler vibrometer corresponds to its surface movement of l/2 = 0.316 mm (half the wavelength of a helium-neon laser source of 0.633 mm). The frequency of this modulation (referred to as the Doppler frequency, Fd) is given by Fd = 2v/l, where v is the surface movement velocity. The recombined light is split into two paths, and a quarter-wave plate is used so that the two signals are in quadrature (sine and cosine) allowing the direction of motion to be determined. A balanced detection scheme, with two detectors in each channel, is used to achieve low noise and high sensitivity. The lock-in amplifier in Fig.10 is a type of amplifier that can extract a signal with a known carrier wave from extremely noisy environment (-60 dB below the noise power). It is, in fact, a homodyne with an extremely low-pass filter (making a very narrow band). Lock-in amplifiers usually measure both in-phase (X) and out-of-phase (Y) components of the signal and can derive the magnitude (R) from them. Many applications of the lock-in technique only require recovering the signal amplitude rather than the relative phase to the reference signal, however here we recover only phase.

Fig.10 2-channel vibrometer (A and B channels are identical except single PZT) The quadrature fringe counting [8] allows combining two distinctive digital phase measuring techniques with mutually complementary characteristics (see Fig.10): one (fast) is counting the Doppler shift frequency, e.g. 8.0 kHz beat frequency for high-velocity measurements, while the other (slow) – the lock-in demodulation with 80 Hz beat frequency for the nm displacement resolution. The two techniques can be operated in switching mode in accordance with the object’s speed in a synchronized way. Two green LEDs are flashed when the demodulator phase goes through 90 and 180 degrees, accordingly. Due to space limitations, Fig.1 has a minor flaw, showing the fringe counter only for B input, but in fact there should be two identical fringe counters. The third rotary knob (Fig.5) noted as “DSP param” serves here as adjuster for either integration time or filter frequency. Several other firmware versions could be developed to maintain required setup and measuring conditions. Please excuse us for omitting Electrochemical Impedances Spectroscopy and many other possible applications. With the new card they appeared pretty trivial.

Conclusions. We have tried to extend cards functionality and usage range as much as possible providing high resolution, low noise, flexibility, etc., nevertheless, there is still place for discussions and improvements. Also microprocessor and digital part schematics are still missing and waits for somebody to get completed.

As the consequence cards concept was leaved open and free. Although the proposed circuits are not properly prototyped, at least an attempt has been made to cope with the mentioned problems, and, possibly all in one solution. ASIO add-ins for MathLab have been found to be quite suitable for scientific research and are preferred host driver, while LabView drivers (National Instruments) are not of primary importance because of SCPI language (www.scpiconsortium.org) speed limitations. The proposed card(s) could make the laboratory measurements in fundamental physics much cheaper, replacing huge amount of expensive scientific rack-mount chassis and boxes.

Acknowledgements The authors wish to thank Dr. L. Latkovskis and Prof. I. Bilinskis for valuable comments and help.

References 1. Mitchell Lee, Understanding and Applying Voltage References, (1999). Linear Technology, Application Note 82. 2. H. Field, UV-VIS-IR Spectral Responsivity Measurement System for Solar Cells, (1998). National Renewable Energy Laboratory, Denver, Colorado report CP-520-25654, www.nrel.gov/docs/fy99osti/25654.pdf; 3. V. Shrotriya, G. Li, Y. Yao, T. Moriarty, K. Emery, Y. Yang, Accurate Measurement and Characterization of Organic Solar Cells, (2006). Advanced Functional Materials, Volume 16 Issue 15, Pages 2016 – 2023; 4. Dubrov M.N., Medvedev P.V. Accurate laser interferometer system for displacement measurements with 1 pm resolution, CAOL 2008, 4th International Conference on Advanced Optoelectronics and Lasers, September 29-October 4, 2008, Alushta, Crimea, Ukraine, p.165-167; [IEEE Cat. No. CFP08814-PRT] 5. M. N. Dubrov and V. A. Alyoshin, Precise laser interferometry with 1 pm resolution (2004), J. of Radioelectronics 5, 2004, http://jre.cplire.ru/jre/may04/1/text.html. 6. Z Yint and M J Usher, A high-resolution wideband digital feedback system for seismometers (1988) J. Phys. E: Sci. Instrum. http://iopscience.iop.org/0022-3735/21/8/002 7. Ometron Laser Doppler Vibrometer — Type 8329, see Fig.1 - Optical Principle of 8329 LDV, Ometron division of Image Automation Ltd. 8. Noh-Bin Yim, Heterodyne Optical Interferometer using Dual Mode Phase Measurement (2001), International Journal of the Korean Society of Precision Engineering, Vol. 2, No. 4. November 2001. 9. Steinberg Media Technologies GmbH, Steinberg Audio Stream Input Output API (2005), www.steinberg.net ,where after registration ASIO SDK will be available for download. 10. Document collection of references above and full size images are available from authors homepage: http://cid-14c8a52578e68092.skydrive.live.com/browse.aspx/Scientific%20DAQ%20USB%20card