LabView Rescues Tiny Signals from a Sea of Noise

Philip F Kromer and Roger Bengtson Department of Physics, University of Texas at Austin Category
R&D / Lab Automation

Products Used
LabView 6.0.2; PCI-MIO-16E-4

The Challenge
Measure an extremely small signal obscured by noise thousands of times greater in magnitude, at a minimum of added expense and hardware, in order to determine the resistance of a high-temperature superconductor.

The Solution
Use the signal-processing capabilities of LabView to implement a technique known as lock-in amplification. Compared to hardware lock-in amplifiers, the LabView approach yields excellent price/performance, increased functionality, superior flexibility, and the ability to inspect the signal at all stages of processing1.

Measuring Small Signals is Difficult

An experiment in our lab determines the resistance of a high-temperature superconductor2 by direct measurement: we apply a known current across the sample, and measure its resistive voltage drop. This signal varies from ~1 mV while normally conducting to <10 nV while superconducting. Even under carefully controlled conditions, there are intractable sources of noise, due to fundamental physical processes, that obscure this signal3. What is worse, a significant portion of this noise has a 1/f spectrum: its intensity increases at low frequencies and is worst at DC, precisely where the signal naturally resides. Finally, intransigent sources of measurement error, such as offset drift, thermoelectric voltages, and common-mode error, act to corrupt the signal. Figure 1 shows a typical input amplitude spectrum (note the log scale). One can see the DC offset, the 60 Hz interference, and the mixture of 1/f and broadband noise. Averaging over a long time reduces the noise by narrowing the bandwidth and effectively trading away response time for improved noise rejection. However, no amount of averaging can distinguish our slowly varying signal from DC and low-frequency components of the noise and error sources.

Figure 1: Log Amplitude versus Frequency (for Fref=1122 Hz)

The Lock-in Technique Provides an Answer

Instead, we will use lock-in amplification4 to recover the signal. Rather than apply a constant (DC) current, we generate a purely sinusoidal (AC) reference signal and apply a scaled current iref = I ref cos( 2 f ref t ) across the sample. The resistive voltage drop has the same frequency and phase as the applied current: vsamp = Vsamp cos( 2 f ref t ) = I ref Rsamp cos( 2 f ref t ) . However, our actual input contains not only the desired vsamp, but also the undesirable offset, interference, and noise (see Figure 1). We can represent the noise and interference a sum of randomly varying signals at all frequencies and phases: vin = vsamp + Voffs + Vn cos( 2 f n t + n ) .

To select only the interesting (matching the reference in phase and frequency) part of the input signal, we use a simple trick from trigonometry. Recall the cosine sum rules: cos( a + b ) = cos( a ) cos( b ) sin( a ) sin( b ) cos( a b) = cos( a ) cos( b) + sin(a ) sin(b ) . Add and rearrange; the product of two cosines yields a cosine at the difference frequency plus a cosine at the sum frequency: 1 cos( a ) cos( b) = [cos( a + b) + cos( a b )] . 2

Multiplying the noisy input signal v in by 2 cos( 2 f ref t ) copies the input, shifted up and down by f ref :
= Vsamp + Vsamp cos( 2 f ref 2 t ) + Voffs cos( f ref 2 t ) + Vn cos([ f n f ref ] 2 t + n ) ,
fn fn

v mult = 2 u ref vin = 2Vsamp cos( 2 f ref t ) cos( 2 f ref t ) + 2Voffs cos( 2 f ref t ) + 2 cos( 2 f ref t ) Vn cos( 2 f n t + n )

The voltage drop has the same frequency as the reference, so their difference term has zero frequency (DC). Therefore, filter out all frequencies below a cutoff frequency f filt << f ref :
Vlock = Vsamp + V cos([ f n f n < f filt n f ref ] 2 t + n ) .

This is the desired voltage drop, with a small remnant of noise, those components indistinguishable in frequency and phase from the reference.

And LabView provides a Solution

Lock-in amplification is traditionally accomplished with expensive, monolithic hardware. Our LabView implementation achieves reasonable performance at greatly reduced price, with superior flexibility5. Besides a National Instruments PCI-MIO-16E-4 data acquisition board6, the only additional hardware devices are an input amplifier and a voltage-controlled current source, both shop-built7. Our LabView program instructs the DAQ board to generate a waveform continuously at the arbitrarily chosen reference frequency of 1122 Hz. This reference drives a voltage-controlled current source connected across the superconductor sample. The superconductors bulk resistance causes a proportional voltage drop, which is passed to a high-gain, low-noise differential amplifier.

Figure 2: Block Diagram for Lock-In Detector

The DAQ board continuously acquires the amplified voltage drop, the voltage from a thermocouple on the sample, and other signals of interest. As each buffer is acquired, our program processes the signals and performs preliminary data analysis. The input signal is locked-in multiplied by the unit reference, scaled, and filtered giving the sample voltage and the nominal bulk resistance (sample voltage over applied current). Since the temperature changes slowly and needs no special attention, we take the average over each buffer (LabView automagically compensates and scales the thermocouple data). The program interactively graphs the various waveforms and their Fourier transforms (see Figure 3). This ability to observe the signal as it proceeds through the lock-in is a distinct advantage of the software approach. Furthermore, the plots display the natural quantities of interest: resistance, time, temperature, etc.; no subsequent data analysis is required. The key to the extraordinary stability and accuracy of our device is that, once acquired, all signal processing is accomplished digitally. Since the reference signal is internally calculated, its accuracy is limited only by the floating-point resolution of the computer. Reading and amplifying the input signal introduces a variety of physical and measurement artifacts, but once safely within the computer no further degradation occurs. There are subtle signal-processing pitfalls that must be avoided, such as synchronization (phase error), aliasing (undersampling) and windowing artifacts (discontinuities at the buffer edges)5. We align acquisition with the waveform generation trigger to minimize phase error, and use continuous acquisition to prevent windowing artifacts. Continuous acquisition allows reference frequencies up to several kilohertz on our modest PC with a 250 kS/s DAQ board. One may instead read non-continuous chunks containing an exact number of waveforms, each synchronous with the waveform generation. This allows reference frequencies up to about ten percent of the maximum sampling frequency.

Figure 3: LabView Front Panel

Results
Figure 4 shows a data set recorded as the sample warmed up from approximately 80 to 200 K using an applied current of 10 mA at 1122 Hz. The input was sampled at 72 kHz, then locked-in and filtered at 0.5 Hz. The inset graph shows the initial portion of the transition on an expanded y-axis. One can clearly observe the shape and details of the transition near 115 K. The transition is not sharp, a general characteristic of high-transition temperature superconductors. Above the transition, we can see that the resistance increases linearly with temperature. Below the transition, we find a superconducting resistance of 0 0.02 microohms.
18

16

Normal conductivity: graph is linear

14

Resistance (milliohms)

12

Onset of superconductivity
Zero-point Resistance
0.08

10

0.04

6
-0.04 80 85 90 95 100 105 110

Conclusions
Using LabView, we have implemented a versatile, low-cost digital lock-in amplifier8. The device 0 50 70 90 110 130 150 170 190 shows negligible offset drift and is robust against Liquid Nitrogen Room Temperature Temperature (K) noise and interference yet it requires minimal Temperature 77 K 300K hardware and may be customized for each task. It Figure 4: Resistance versus Temperature for a High-Temperature Superconductor is capable of 10 nV sensitivity, a quality factor of 5 Q = ? f / f = 10 or more, and noise rejection of ~120 dB (can extract signals from noise up to ~106 times greater in amplitude). For implementation details, circuit diagrams, and source code, please see http://www.ph.utexas.edu/~phy453/lockin/ or contact the authors: Philip (Flip) Kromer (flip@physics.utexas.edu) or Roger Bengtson (bengtson@physics.utexas.edu). 2 Superconductivity: G.C. Brown, J.O. Rasure, and W.A. Morrison, American Journal of Physics. 57(12), 1142-1144 (1989). M.J. Pechan and J.A. Horvath, American Journal of Physics.58(7), 642-644 (1990). Semiconductor kits are available from Colorado Superconductor, 1623 Hillside Drive, Fort Collins, CO 80524. 3 Sources of, and defenses against, noise: "Signal Enhancement" (http://www.srsys.com/html/applicationnotes.html, or p.225 of their catalog). Stanford Research Systems, Sunnyvale, CA, 1999. A summary of fundamental noise sources. S.J. Shah, Field Wiring and Noise Considerations, National Instruments, Austin, TX, 1994; see http://digital.natinst.com/appnotes.nsf/web/index, #25. Low Level Measurements Handbook, ed. J. Yeager and M.A. Hrusch-Tupta. Keithley Instruments, Cleveland, OH, 1998. An excellent introduction to precision measurement, and freely available upon request. P. Horowitz and W. Hill, The Art of Electronics. Cambridge University Press, New York, 1980. 4 Lock-in detection: M. Stachel, "The Lock-in Amplifier: Exploring Noise Reduction and Phase," (http://www.lockin.de/). An excellent webbased introduction to lock-in detection, complete with Java simulations. P. Temple, American Journal of Physics 43(9), p801 (1975). "About Lock-in Amplifiers" (http://www.srsys.com/html/applicationnotes.html). Stanford Research Systems, Sunnyvale, CA, 1999. A functional description of lock-in amplifiers. Lock-in Applications Anthology, ed. Douglas Malchow. EG&G Princeton Applied Research, Princeton, NJ, 1985. A freely available guide to applications of the lock-in analyzer. D.W. Preston and E.R. Dietz, The Art of Experimental Physics, pp 367-375. John Wiley & Sons, New York, 1991. 5 Data Acquisition: Data Acquisition Handbook, ed. J. Yeager and M.A. Hrusch-Tupta. Keithley Instruments, Cleveland, OH, 1998. An excellent introduction to data acquisition, and freely available on request. 6 We use the PCI-MIO-16E-4 (NI 6040E) multifunction I/O board; it has 16 12-bit, 250 kS/s analog inputs; two 12-bit, 1 MS/s analog outputs; and two 24-bit counters. National Instruments, 11500 N. MoPac Expressway, Austin, TX 78759 7 Our front-end amplifier is based on Texas Instruments' INA114 precision instrumentation amplifier. Other suitable devices include Analog Devices' AD624 and Texas Instruments' OPA111. A circuit diagram is available on our website. 8 Our source code is freely available under the Gnu Public License; download at http://www.ph.utexas.edu/~phy453/lockin/.
1
2