Super-resolution imaging of a low frequency levitated oscillator

Super-resolution imaging of a low frequency levitated oscillator

N. P. Bullier Department of Physics and Astronomy, University College London, Gower Street, London WC1E 6BT, United Kingdom    A. Pontin Department of Physics and Astronomy, University College London, Gower Street, London WC1E 6BT, United Kingdom    P. F. Barker Department of Physics and Astronomy, University College London, Gower Street, London WC1E 6BT, United Kingdom

We describe the measurement of the secular motion of a levitated nanoparticle in a Paul trap with a CMOS camera. This simple method enables us to reach signal-to-noise ratios as good as 10 with a displacement sensitivity better than 10 m/Hz. This method can be used to extract trap parameters as well as the properties of the levitated particles. We demonstrate continuous monitoring of the particle dynamics on timescales of the order of weeks. We show that by using the improvement given by super-resolution imaging, a significant reduction in the noise floor can be attained, with an increase in the bandwidth of the force sensitivity. This approach represents a competitive alternative to standard optical detection for a range of low frequency oscillators where low optical powers are required.

I Introduction

Levitated nanoparticle oscillators are seen as promising platforms for exploring the macroscopic limits of quantum mechanics. For example, levitated nanoparticles containing a single spin are potential candidates for creating a macroscopic superposition via a type of Ramsey interferometry Sougato1 (). Motional cooling of nanoparticles within a trap, followed by a controlled release through a matter wave diffraction grating, also offers the potential for creating and evidencing macroscopic superposition Bateman2014 (). More recently, non-interferometric methods have been proposed as a way of testing wavefunction collapse models by measuring the excess noise in the trapped motion of a levitated nanoparticle Goldwater ().

A distinguishing feature of many of these levitated oscillators is low oscillation frequency, typically in the range of a few Hz to kHz. Detection of small amplitude mechanical motion at these low frequencies is challenging as environmental mechanical noise in the detection chain can swamp measurements of the trapped motion of the nanoparticle. A common approach to measurement of particle displacement uses a difference detection scheme where a laser beam is used to illuminate the particle. The transmitted light modified by scattering from the particle is typically directed onto a quadrant photodiode or alternatively split into two equal components using a mirror such that each component is detected on a separate photodiode. The motion of the particle is observed as an imbalance in the difference between the currents on the photodiodes Gieseler (). While this is relatively noise free at frequencies exceeding 10 kHz, low frequency mechanical noise can induce beam pointing noise which cannot easily be differentiated from the true mechanical motion of the oscillator.

We present a very simple and surprisingly sensitive imaging method for measuring low frequency particle displacement which is free of the low mechanical noise observed in split detection methods but yet demonstrates displacement sensitivity of better than 10 m/Hz.

Figure 1: (a) An image of the trap with a nanoparticle in its centre. Image taken from the camera axis (y-axis). The end of the quadrupole guide can be seen on the right-hand side of the picture. It is used to load the trap and as a quadrupole mass filter to select a given charge-to-mass ratio. (b) A schematic view with camera position and axes defined. The laser propagates along the x-axis. The trap parameters discussed in the main text are the following:  mm,  mm, (asymmetric driving), , and .

Ii Levitation in a Paul trap

We use a single nanoparticle trapped in a linear Paul trap to create a levitated nanoparticle oscillator whose trap frequency is dependent on the surface charge and the applied voltage on the electrodes of the trap. A picture of the trap is shown in Fig. 1(a) which consists of four electrodes placed at the corner of a square to provide an AC field that confines the nanoparticle in the x-y plane with two additional electrodes to apply a DC voltage that traps the motion along the z-axis. In this configuration, the total electric field near the center of the trap can be written as wineland ():


where and are the distances between the trap center and the AC and DC electrodes respectively, and are the applied potentials, is the AC drive frequency, and are geometrical efficiency coefficients efficiency1 (); efficiency2 () quantifying non-perfectly quadratic potentials and is the trap ellipticity. We obtain the efficiency parameters with numerical FEM simulations (see caption of Fig. 1 for the numerical values). We drive the trap asymmetrically by applying the same AC potential on two opposite electrodes and grounding the other two.

The AC potential gives rise to a pseudo-potential with and the charge-to-mass ratio. Following from the expression of the electric field given in Eq. 1, one can define stability parameters wineland () and corresponding to the motion along the -axis given by the AC and DC potential respectively:


where we assumed . In the case where , , the pseudo-potential gives rise to secular frequencies which can be approximated to .

The trap is loaded with nanoparticles in low vacuum, at a pressure of  mbar, by means of electrospray ionization electrospray (), where the nanoparticles are suspended in ethanol with a concentration  10 g/mL. In order to increase the flux of particles reaching the trapping region, a quadrupole guide “connects” the output skimmer and the Paul trap which are placed roughly  cm apart. Its end section can be seen in Fig. 1(a). The guide is usually operated in a mass filter configuration march (), where a high DC voltage is applied onto two electrodes such that only the tip of the first stability region of the guide is allowed. Note that since the trap and the guide have different geometries, the effective stability region is further reduced. We typically trap particles with charge-to-mass ratios in the range  C/kg.

Iii Measuring displacement via imaging

Optical interferometric detection schemes often offer the best sensitivities in terms of displacement. However, typical sensitivities are often much worst at low frequency. For example, the noise floor of a standard balanced Michelson interferometer can be of the order of  m/Hz around  kHz but it can easily drop to  m/Hz at  Hz due to environmental noise affecting the beam paths as well as flicker noise Slezak2018 (). This difference can be even more pronounced when dealing with nanoparticle motion due to the intrinsically lower coupling/detection efficiency.

Figure 2: Time-traces of the motion along the z-axis of a 200 nm (radius) silica nanosphere at 10 mbar, derived from imaging. The particle position is obtained by considering the pixel of highest intensity. The displacement is calibrated by moving the camera on a translational stage. The resulting calibration coefficient is (3.64m/pixel.

Here, we exploit a much simpler detection scheme. The trapped nanoparticle is illuminated with a green laser diode with powers ranging from 10 mW to 40 mW and a beam waist of 250 m. The light scattered at °  is collected by a zoom objective objective () mounted on a lens lens () and imaged by a low cost CMOS camera camera (). We extract the particle position on the x-z plane by finding the coordinates of the pixel of highest intensity. The motion of the particle is calibrated by moving the camera by a known amount using a translational stage. An example of a time trace for a silica nanosphere of 200 nm of radius can be seen in Fig. 2. The camera sensor has a resolution of 1280 x 1024, however a smaller area of interest can be addressed, allowing an acquisition at a faster frame rate. In our typical configuration we exploit a 50 x 40 matrix that allows to acquire at  fps. This, in combination with the simplicity of the algorithm to reconstruct the particle position, allows for a real time acquisition of two time traces corresponding to the particle motion projected onto the coordinates defined by the camera pixel matrix. It is worth noting that the time series are composed of  bit integers. Combined with the low sampling rate allows for continuous monitoring of the particle position over very long times without having large data storage requirements.

The motion of the nanoparticle in the trap can be decomposed in two different motions. A secular motion, driven by stochastic forces, such as collisions by gas particles, and a motion at the drive frequency . Here, typically, is a few kHz. The secular motion can be easily kept at frequencies smaller than  Hz. The motion can be described by a damped harmonic oscillator. The single-sided power spectral density (PSD) of the motion along any axis i can be written as: where is the PSD of the thermal Langevin force while is the PSD of the total force noise due to all other unknown sources. The bath temperature is given by , is the Boltzmann constant, the mass of the nanoparticle and the viscous damping due to collisions with residual gas. Finally, , is the mechanical susceptibility, given by where is the resonant frequency which here corresponds to the secular frequency along the i axis.

We show in Fig. 3 typical spectra for the same nanoparticle as the one shown in Fig. 2 taken at different pressures along the z-axis. Despite the very simple approach, this method provides spectra with high signal-to-noise ratios (SNR). Along with the experimental spectra, we show fit results assuming a total force noise PSD. Down to pressures of the order of  mbar the spectral noise floor are sufficiently low to resolve the susceptibility down to the zero frequency limit.

Figure 3: Calibrated Power Spectral Densities (PSD) of the motion of a trapped nanoparticle at 0.10 mbar and 1.010 mbar. Blue continuous lines are experimental PSD of the motion along the z-axis. The secular frequency is 92.5 Hz. The broader linewidth corresponds to the motion at 0.10 mbar. The fits of the mechanical susceptibilities are shown in dashed cyan. The time trace shown in Fig. 2 is a subsection of the timetrace used to obtain the spectra shown at 1.010 mbar.

The continuous monitoring of the particle motion, allows us to explore the dynamics down to extremely low frequencies. Taking into consideration the longest continuous stretch of data that lasts almost  days we obtain the displacement PSD shown in Fig. 4. Interestingly, a peak (and its harmonics) is clearly visible corresponding to a period of  hour. It is possible to show that this displacement modulation is highly correlated to a modulation of the secular frequency, itself correlated to the temperature drifts in the lab. Approximately 25 % of this drift can be accounted for in drifts of the amplitude of the voltage used to drive the Paul trap. The amplitude of the signal generator changes by 1.2 mV (amplitude peak-to-peak) which corresponds after amplification to 240 mV (amplitude peak-peak) on the trap electrodes. The other 75 % is assumed to come from the deformation of the trap itself with temperature.

Figure 4: Low frequency PSDs calculated from data acquired for almost three days. Red and blue lines represent the PSD along the x and z-axis respectively. The peak around  hour corresponds to motion in the trap correlated to temperature fluctuations in the lab. At frequencies above 10 Hz, the PSD has been averaged with more spectra, which leads to the smoother PSD profiles in this range.

Iv Super-resolution measurements

The spectra shown so far in Figures 3 and 4 have been calculated on time traces where the position of the particle has been determined by finding the pixel of highest intensity. This naive approach enables real time traces by controlling the camera with a Python library instrumental (). This is enough to extract basic information such as frequencies and linewidths characterizing the mechanical motion. Applications such as sensing or cooling of the centre-of-mass motion may require higher sensitivities which can be achieved by using super-resolution imaging.

Complex algorithms to obtain sub-pixel resolution is applied in many fields including particle physics subpx1 (), chemistry subpx2 () and astrophysics. In the latter case, for example, it was applied to enhance the search for exoplanets subpx3 (). Here, we extract the particle position by fitting a two-dimensional Gaussian profile to the spatial intensity distribution given by the camera. This increases the sensitivity by more than two orders of magnitude as it can be seen in Fig. 5, where we plot a comparison between PSDs with and without the resolution enhancement for both trap directions. These spectra were taken at a pressure  mbar for a particle with a charge-to-mass ratio of  C/kg. The secular frequencies along the x, y, and z-axis are , and  Hz respectively. A trap ellipticity slightly different from the nominal one of 0.5 removes the degeneracy on the x and y-axes, which are easily resolved at this pressure. The peak in the PSD at  Hz corresponds to the aliasing of the excess micromotion at the AC drive frequency of  kHz.

Figure 5: Resolution enhancement obtained by fitting a Gaussian distribution to find the particle position on each frame. (a) Time-trace of the motion along the z-axis obtained with the resolution enhancement showing sub-pixel resolution. (b) PSD of the particle displacement. Red and blue lines refer to the motion along the z and x-axis respectively. The gray shaded area shows the noise floor when the algorithm finds the intensity maximum to obtain the particle position.

Given its low mass and small damping rate, a levitated nanosphere in vacuum is a sensitive force sensor. The super-resolution enhancement can be used to increase the force sensitivity bandwidth too. We show an example in Fig. 6 that compares the force sensitivity of a 200 nm radius silica nanosphere levitated at 7.410 mbar using super-resolution and the method that finds the brightest pixel to extract the particle position. It enables us to increase the 3 dB force-noise sensitivity bandwidth from 21 Hz to 112 Hz. This last one peaks at (1.00.1) aN. At this pressure, the force noise contribution from the Brownian motion, and therefore the lowest achievable limit without other noise, is =(2.30.2) zN. The force noise sensitivity, , shown in Fig. 6 is calculated from the displacement PSD as with the mechanical susceptibility defined above. The mechanical susceptibility depends on the mass, independently measured, as well as the damping and the secular frequency, obtained by fitting a displacement PSD from an independent data set. This approach provides a good estimate of the force sensitivity, however, a more rigorous analysis can be obtained exploiting Wiener filter theory wienerfiltering ().

Figure 6: Force-noise sensitivity of a 200 nm silica radius nanosphere at 7.410 mbar along the z-axis. The red curve corresponds to the force-noise sensitivity obtained without the super-resolution enhancement. The blue one is obtained on the same acquisition but using the super-resolution. This increases the 3 dB bandwidth, centred around 90 Hz, from 21 Hz to 112 Hz.

V Conclusions

We have demonstrated the use of a low-cost CMOS camera to make precision displacement measurements of a low frequency high-Q mechanical oscillator. Although applied to a Paul trap, this method could be used for many applications from sensing to cooling in a range of oscillators.

The three factors which can be improved in the measurements presented here are the sensitivity, the acquisition rate and the fitting speed of the algorithm used to determine the particle position. The sensitivity of the measurements can easily be improved by increasing the magnification. The secular frequencies were limited here to 500 Hz by the acquisition rate. In order to directly detect higher secular frequencies and avoid aliasing, it is possible to use cameras with higher pixel clocks (i.e.,  MHz against  MHz here), opening the possibility of getting frame rates as high as  kHz. Furthermore, fluorescence correlation spectroscopy techniques Kim2007 () could be used in the case of fluorescing nanoparticles such as YLF crystals doped in Yb refrigiration (). This could lead to frame rates of tens of kHz. Recent Single-Photon Avalanche Diode (SPAD) cameras have been operated up to  kfps spad ().

While here the time traces without subpixel resolution were live, the data with super-resolution were obtained by post-processing the video. Indeed, the fitting routine used was too slow to keep acquiring data at  frames per second (fps). It is nonetheless still possible to acquire live position time traces with very high frame rates by running the fitting algorithm on a Graphics Processing Unit (GPU). Gaussian fits or better algorithms can then achieve fit speeds  fits/s for a 10 x 10 pixel matrix fastfit1 (); fastfit2 ().

Vi acknowledgments

The authors acknowledge funding from the EPSRC Grant No. EP/N031105/1 and from the EU H2020 FET project TEQ (Grant No. 766900). NPB acknowledges funding from the EPSRC Grant No. EP/L015242/1. AP has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Grant Agreement No. 749709.


  • [1] Sougato Bose, Anupam Mazumdar, Gavin W. Morley, Hendrik Ulbricht, Marko Toroš, Mauro Paternostro, Andrew A. Geraci, Peter F. Barker, M. S. Kim, and Gerard Milburn. Spin entanglement witness for quantum gravity. Phys. Rev. Lett., 119:240401, Dec 2017.
  • [2] James Bateman, Stefan Nimmrichter, Klaus Hornberger, and Hendrik Ulbricht. Near-field interferometry of a free-falling nanoparticle from a point-like source. Nature Communications, 5(1), sep 2014.
  • [3] Daniel Goldwater, Mauro Paternostro, and P. F. Barker. Testing wave-function-collapse models using parametric heating of a trapped nanosphere. Phys. Rev. A, 94:010104, Jul 2016.
  • [4] Jan Gieseler, Bradley Deutsch, Romain Quidant, and Lukas Novotny. Subkelvin parametric feedback cooling of a laser-trapped nanoparticle. Phys. Rev. Lett., 109:103603, Sep 2012.
  • [5] D. J. Berkeland, J. D. Miller, J. C. Bergquist, W. M. Itano, and D. J. Wineland. Minimization of ion micromotion in a paul trap. Journal of Applied Physics, 83(10):5025–5033, 1998.
  • [6] R. R. A. Syms, T. J. Tate, M. M. Ahmad, and S. Taylor. Design of a microengineered electrostatic quadrupole lens. IEEE Transactions on Electron Devices, 45(11):2304–2311, Nov 1998.
  • [7] M.J. Madsen, W.K. Hensinger, D. Stick, J.A. Rabchuk, and C. Monroe. Planar ion trap geometry for microfabrication. Applied Physics B, 78(5):639–651, Mar 2004.
  • [8] John B. Fenn, Matthias Mann, Chin Kai Meng, Shek Fu Wong, and Craige M. Whitehouse. Electrospray ionization-principles and practice. Mass Spectrometry Reviews, 9(1):37–70, 1990.
  • [9] Raymond E. March. Quadrupole ion traps. Mass Spectrometry Reviews, 28(6):961–989, 2009.
  • [10] Bradley R Slezak, Charles W Lewandowski, Jen-Feng Hsu, and Brian D’Urso. Cooling the motion of a silica microsphere in a magneto-gravitational trap in ultra-high vacuum. New Journal of Physics, 20(6):063028, jun 2018.
  • [11] The zoom objective used is a Zoom 7000 by Navitar.
  • [12] The lens used is a fixed focal length Lens extender from Edmund optics.
  • [13] The camera used is a by ThorLabs.
  • [14]
  • [15] T. Bisanz, J. Große-Knetter, A. Quadt, J. Rieger, and J. Weingarten. Subpixel mapping and test beam studies with a HV2fei4v2 CMOS-sensor-hybrid module for the ATLAS inner detector upgrade. Journal of Instrumentation, 12(08):P08018–P08018, aug 2017.
  • [16] Stanisław H. Nowak, Aniouar Bjeoumikhov, Johannes von Borany, Josef Buchriegler, Frans Munnik, Marko Petric, Martin Radtke, Axel D. Renno, Uwe Reinholz, Oliver Scharf, and Reiner Wedell. Sub-pixel resolution with a color x-ray camera. Journal of Analytical Atomic Spectrometry, 30(9):1890–1897, 2015.
  • [17] Stephen T. Bryson, Peter Tenenbaum, Jon M. Jenkins, Hema Chandrasekaran, Todd Klaus, Douglas A. Caldwell, Ronald L. Gilliland, Michael R. Haas, Jessie L. Dotson, David G. Koch, and William J. Borucki. The Kepler pixel response function. The Astrophysical Journal, 713(2):L97–L102, mar 2010.
  • [18] A. Pontin, M. Bonaldi, A. Borrielli, F. S. Cataliotti, F. Marino, G. A. Prodi, E. Serra, and F. Marin. Detection of weak stochastic forces in a parametrically stabilized micro-optomechanical system. Phys. Rev. A, 89:023848, Feb 2014.
  • [19] Sally A Kim, Katrin G Heinze, and Petra Schwille. Fluorescence correlation spectroscopy in living cells. Nature Methods, 4(11):963–973, October 2007.
  • [20] A. T. M. Anishur Rahman and P. F. Barker. Laser refrigeration, alignment and rotation of levitated yb3+:ylf nanocrystals. Nature Photonics, 11(10):634–638, sep 2017.
  • [21] Genevieve Gariepy, Nikola Krstajić, Robert Henderson, Chunyong Li, Robert R. Thomson, Gerald S. Buller, Barmak Heshmat, Ramesh Raskar, Jonathan Leach, and Daniele Faccio. Single-photon sensitive light-in-fight imaging. Nature Communications, 6(1), jan 2015.
  • [22] Yiming Li, Markus Mund, Philipp Hoess, Joran Deschamps, Ulf Matti, Bianca Nijmeijer, Vilma Jimenez Sabinina, Jan Ellenberg, Ingmar Schoen, and Jonas Ries. Real-time 3d single-molecule localization using experimental point spread functions. Nature Methods, 15(5):367–369, apr 2018.
  • [23] Carlas S Smith, Nikolai Joseph, Bernd Rieger, and Keith A Lidke. Fast, single-molecule localization that achieves theoretically minimum uncertainty. Nature Methods, 7(5):373–375, apr 2010.
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
Add comment
Loading ...
This is a comment super asjknd jkasnjk adsnkj
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test description