# An inverse method for color uniformity in white LED spotlights

- Corien Prins
^{1}Email author, - Jan ten Thije Boonkkamp
^{1}, - Teus Tukker
^{2}and - Wilbert IJzerman
^{3}

**4**:5

https://doi.org/10.1186/2190-5983-4-5

© Prins et al.; licensee Springer. 2014

**Received: **13 December 2012

**Accepted: **24 March 2014

**Published: **3 June 2014

## Abstract

Color over Angle (CoA) variation in the light output of white phosphor-converted LEDs is a common problem in LED lighting technology. In this article we propose an inverse method to design an optical element that eliminates the color variation for a point light source. The method in this article is an improved version of an earlier method by the same authors, and provides more design freedom than the original method. We derive a mathematical model for color mixing in a collimator and present a numerical algorithm to solve it. We verify the results using Monte-Carlo ray tracing.

## 1 Introduction

LED is a rising technology in the field of lighting. In the past, LEDs were only suitable as indicator lights, but the enormous improvements in energy efficiency, cost and light output now allow the use of LEDs for lighting applications [1]. Additionally, LED lighting benefits from low maintenance cost and long lifetime.

Because LED is a rising technology, companies and researchers are constantly searching for methods to reduce the production cost and increase the efficiency, light output and light quality of LED-based lamps. An important issue for white LED lamps is color variation of the emitted light. This is caused by color variation in the light output of the most common type of white LED, the phosphor-converted LED. This type of LED consists of a blue LED with on top a so-called phosphor layer which converts part of the blue light into yellow and red. The resulting output is white light. The distance that a light ray travels through the phosphor depends on the angle of emission. As a result, the light emitted normal to the LED surface is more bluish, while the light emitted nearly parallel to the surface is more yellowish [[2], pp.353-357]. This phenomenon is called Color over Angle (CoA) variation.

A lot of research has been done to reduce this color variation. Introduction of bubbles in the phosphor layer causes scattering of light, reducing the color variation [3]. Another common method is the application of a dichroic coating on the LED [4]. However, these methods reduce the efficiency of the LED and increase the production costs. Wang *et al*. [5] proposed a modification of the optics on the LED to improve the color uniformity. In the case of a spot light, the LED is combined with a collimator. A collimator is an optical component that reduces the angular width of the light emitted by the LED. A common technique is to add a microstructure on top of the collimator. However, this microstructure introduces extra costs in the production process of the collimator, makes the collimator look unattractive and broadens the light beam.

None of the methods mentioned above rigorously solve the problem of color variation, and all methods reduce the efficiency of the optical system. In earlier work [6], we introduced an inverse method to design a specific type of collimator, the so-called TIR (total internal reflection) collimator. The TIR collimator designed with this method mixes light from a point source such that the color variation is completely eliminated. The collimator requires no microstructures nor scattering techniques. However, the inverse method left very little design freedom for optical designers. An optical designer wants to influence the height and width of the collimator, for example, to fit it into the available space in a lamp design. Also, optical designers want a color mixing collimator which resembles a standard collimator as closely as possible. The inverse method introduced in this paper is an improvement of the method introduced in [6]. The collimator has three free surfaces instead of two. As a result, the improved method offers more design freedom, and it is nearly impossible to distinguish the resulting collimator with the naked eye from a collimator without color correction.

The contents of this paper is the following. First we give a thorough introduction to inverse methods for optical systems and the theory of color mixing in Section 2. In Section 3 we explain the improved inverse method. Section 4 describes three examples where the new method is used. Finally, we end with concluding remarks in Section 5.

## 2 Design of a TIR collimator using inverse methods

*t*of rays leaving the LED and the angles

*θ*of rays leaving the collimator, the so-called transfer functions. Subsequently we use these transfer functions to calculate the free surfaces of the TIR collimator such that the light is redirected according to the relation defined by the transfer functions. In Figure 1 these free surfaces are denoted by A, B and C.

### 2.1 Source and target intensities

*z*-axis (inclination), and $u\in [0,2\pi )$ is the angle that rotates around the

*z*-axis (azimuth). Because of the symmetry of the system, the intensity $I(t,u)$ is independent of

*u*and denoted by $I(t)$. We introduce an effective intensity $\mathcal{I}(t)$, which is the flux per rad through the circular strip $[t,t+\mathrm{d}t]$ on the unit sphere divided by 2

*π*. We calculate $\mathcal{I}(t)$ by integrating $I(t)$ over the angle

*u*:

The effective intensity has unit [lm/rad]. For an LED, the effective intensity is typically positive for $t\in (0,\pi /2)$.

*ϕ*. Integration over the angle

*ϕ*results in an effective intensity $\mathcal{G}(\theta )=sin(\theta )G(\theta )$ [lm/rad]. A more in-depth discussion of effective intensity distributions can be found in Maes [7]. The target intensity is multiplied by a constant $c>0$ such that we have conservation of luminous flux for the optical system:

*θ*, ${\eta}_{i}(\theta )$ gives the emission angle

*t*of the LED in $[{\tau}_{i-1},{\tau}_{i}]$. We choose each transfer function to be strictly monotonic and thus invertible. The luminous flux emitted from the collimator in the interval $[\theta ,\theta +\mathrm{d}\theta ]$ must be equal to the sum over

*i*of the luminous fluxes emitted from the source in each interval $[{\eta}_{i}(\theta ),{\eta}_{i}(\theta +\mathrm{d}\theta )]$. This leads to the following relation:

where ${\sigma}_{i}=-1$ for monotonically decreasing transfer functions and ${\sigma}_{i}=1$ for monotonically increasing transfer functions.

### 2.2 Color mixing

The second requirement on transfer functions is related to the color of the resulting beam from the collimator. First we give a short introduction to the theory of color perception, then we derive an ordinary differential equation describing the color of the beam.

The resulting chromaticity coordinates are weighted averages of the chromaticity coordinates of the original beam with weights ${L}_{1}/{y}_{1}$ and ${L}_{2}/{y}_{2}$. Note that a point $(x,y)$ is on the straight line segment between $({x}_{1},{y}_{1})$ and $({x}_{2},{y}_{2})$.

*t*and are described by functions $x(t)$ and $y(t)$. From measured data we have observed an approximate linear relationship between $x(t)$ and $y(t)$, see Figure 2. From the color mixing rule, we conclude that if we mix light from different angles of the LED into a single beam with color coordinates $({x}_{T},{y}_{T})$, these coordinates must be on the straight line segment relating $x(t)$ and $y(t)$. Therefore, given ${y}_{T}$, the chromaticity coordinate ${x}_{T}$ is fully determined and we only need that the

*y*-coordinate of the mixed light equals a certain constant target value ${y}_{T}$. The light in the interval $[\theta ,\theta +\mathrm{d}\theta ]$ emitted from the TIR collimator is the sum of beams with intensity ${\sigma}_{i}\mathcal{I}({\eta}_{i}(\theta ))\phantom{\rule{0.2em}{0ex}}\mathrm{d}{\eta}_{i}(\theta )$. The

*y*-coordinate of this light is therefore

### 2.3 Free surface calculations

*S*from which the rays depart with a given angle $\tilde{t}$:

*f*is the distance a light ray travels from the surface

*S*to the free surface,

*s*is the arc-length along

*S*, $\tilde{t}$ is the angle of the ray leaving

*S*with respect to the

*z*-axis, and

*δ*is the ray-emission angle measured counterclockwise with respect to the normal of

*S*. The angle

*β*is the angle of incidence on the free surface with respect to the surface normal. For the reflective surfaces B and C, the variables are illustrated in Figure 3. We like to formulate this differential equation in terms of

*t*instead of

*s*. Multiplication by $\mathrm{d}s/\mathrm{d}t$ gives

*β*and

*δ*depend on

*t*and are derived below.

Light propagates through the collimator by two type of routes. In the ‘TIR route’, light is refracted by surface *S*, reflected by surface B or C by total internal reflection and finally refracted by surface *T*. In the ‘lens route’, light is refracted by surface A and subsequently refracted by surface *T*.

*S*. Let

*d*be the distance from the left of surface

*S*to the LED and

*α*the clockwise angle of this surface with respect to the symmetry-axis. A ray that leaves the LED at angle

*t*, will hit surface

*S*at $({x}_{S}(t),{y}_{S}(t))$ and leave surface

*S*with angle $\tilde{t}(t)$ given by

*n*. From (9a) and (9b) we find

*s*, which is defined to be 0 at $t=\pi /2$. Also we calculate

*δ*:

*z*-axis after reflection. Before the rays leave the TIR collimator, they are once more refracted by surface

*T*. Rays that leave the collimator at angle $\theta ={\eta}^{-1}(t)$ must enter surface

*T*at angle

*z*-axis, and positive otherwise. Equation (12) is derived using Snell’s law. Now we can calculate $f(t)$ by numerically integrating the ODE (8) backwards, starting at $t=\pi /2$. The parameters in (8) are given by (9c), (10a), (10b), (11) and (12). For surface C, a plus sign is chosen in (12) and for surface B a minus sign. The integration for surface C starts with $f(\pi /2)=b\ge 0$, which is usually chosen larger than 0 to prevent a sharp edge of the collimator for manufacturing purposes. At $t={\tau}_{2}$, the final value $f({\tau}_{2})$ of the calculation of surface C is chosen as starting value for the calculation of surface B. The coordinates of the surfaces B and C can be calculated as

*β*[11]:

## 3 A TIR collimator with three transfer functions

*t*increases with

*θ*, so ${\sigma}_{1}={\sigma}_{3}=1$. For surface B,

*t*decreases for increasing values of

*θ*, and thus ${\sigma}_{2}=-1$. We use the following convention: ${\mathcal{I}}_{i}(\theta )=\mathcal{I}({\eta}_{i}(\theta ))$ and ${y}_{i}(\theta )=y({\eta}_{i}(\theta ))$. We now have the following system of differential equations:

*j*be the index of this transfer function, then we impose

The system (22a)-(22c) with boundary conditions (19a) and (19b) has three unknown functions ${\eta}_{1}(\theta )$, ${\eta}_{2}(\theta )$ and ${\eta}_{3}(\theta )$. The functions $\mathcal{I}(t)$ and $y(t)$ are known from measurements on the LED. The function $\mathcal{G}(\theta )$ can be chosen by the optical designer as a finite function on $[0,{\theta}_{max}]$. The constants *c*, ${y}_{T}$ and *r* cannot be chosen freely, we will show that their values follow from conservation of luminous flux, the law of color mixing and the choice of ${\tau}_{1}$ and ${\tau}_{2}$, respectively. Also the values of ${\tau}_{1}$ and ${\tau}_{2}$ cannot be chosen freely, we will derive an inequality that guarantees monotonicity of the transfer functions.

We choose the positive sign in front of the square root since ${\eta}_{1}^{\prime}(0)$ should be positive. Here ${\mathcal{G}}_{+}^{\prime}(0)$ and ${\mathcal{I}}_{+}^{\prime}(0)$ are the right derivatives of $\mathcal{G}(\theta )$ at $\theta =0$ and of $\mathcal{I}(t)$ at $t=0$, respectively. These right derivatives are positive because $\mathcal{I}(t)$ and $\mathcal{G}(\theta )$ are positive at $t>0$ and $\theta >0$. We have $y(t)>0$ by definition of chromaticity coordinates, and we assume based on measurements that $y({\tau}_{1})<y({\tau}_{2})$ and thus ${y}_{1}(0)<{y}_{3}(0)$. From this we see that we need to choose ${\tau}_{2}$ such that $y({\tau}_{2})>{y}_{T}$, so the right hand side of (23) is positive and real.

### 3.1 The values of *c*, ${y}_{T}$and *r*

The system (21) with boundary conditions (19a) and (19b) appears to be overdetermined. However, the system contains three unknown parameters which still need to be chosen. We derive values for three constants *c*, ${y}_{T}$ and *r* given the boundary conditions and assuming monotonicity of the transfer functions. Later we show that our choice of the constants *c*, ${y}_{T}$ and *r* imply that three of the boundary conditions are superfluous.

*c*. Integration of the first row of (21), using the given boundary conditions and ${\mathcal{I}}_{i}(\theta )=\mathcal{I}({\eta}_{i}(\theta ))$, yields

The function $\mathcal{I}(t)$ is known from measurements on the LED, the function $\mathcal{G}(\theta )$ is chosen by the optical designer, so from this relation we derive the value of the constant *c*. This relation corresponds to conservation of luminous flux (equation (2)).

This relation shows that ${y}_{T}$ is the weighted harmonic average of the *y*-chromaticity coordinate of the light source. Like $\mathcal{I}(t)$, the function $y(t)$ is known from measurements on the LED, thus we can derive the value of ${y}_{T}$.

*r*. Integration of the third row of (21) with the given boundary conditions and substitution of (24) yields

This relations corresponds to conservation of luminous flux for the second transfer function.

### 3.2 Monotonicity of the transfer functions

The transfer functions calculated from (22a)-(22c) should be monotonic, otherwise they have no physical meaning. From (22b) we can easily see that ${\eta}_{2}^{\prime}(\theta )\le 0$ because $r>0$, $\mathcal{G}(\theta )\ge 0$ and $\mathcal{I}(t)\ge 0$, thus, ${\eta}_{2}(\theta )$ is monotonically decreasing. The monotonicity of ${\eta}_{1}$ and ${\eta}_{3}$ is more complicated to show and we need some additional assumptions to derive a sufficient condition for monotonicity.

**Theorem 1**

*Assume that the chromaticity coordinate function*$y(t)$

*satisfies the inequalities*

*and the transfer functions satisfy the bounds*

*If*

*then* ${\eta}_{1}(\theta )$ *and* ${\eta}_{3}(\theta )$ *are monotonically increasing*.

*Proof*We need to prove that the derivatives of ${\eta}_{1}$ and ${\eta}_{3}$ are positive. From (22a), using assumptions (27) and (28), we find that ${\eta}_{1}$ is monotonically increasing if

The function $M(\theta )$ it is monotonically increasing because ${\eta}_{2}(\theta )$ is monotonically decreasing. Therefore, if $y({\tau}_{1})\le M(0)$, we find using (27) and (28) that ${y}_{1}(\theta )\le M(\theta )$ for all *θ*. Similarly, if $M({\theta}_{max})\le y({\tau}_{2})$, then $M(\theta )\le {y}_{3}(\theta )$ for all *θ*. The inequalities $y({\tau}_{1})\le M(0)$ and $M({\theta}_{max})\le y({\tau}_{2})$ are equivalent to the second and third inequality in (29). □

### 3.3 The initial value problem

The ODE-system (21) with the boundary conditions (19a) and (19b) can be solved as an initial value problem. We remark that solving the system as an end value problem has no advantages or disadvantages. We discard the end conditions and solve the initial value problem using a Runge-Kutta method. The end conditions are satisfied as a result of our choices of *c*, ${y}_{T}$ and *r*.

**Theorem 2** *Assume monotonicity of the transfer functions*. *The solution of the initial value problem defined by the ODE system* (21) *and the initial conditions* ${\eta}_{1}(0)=0$, ${\eta}_{2}(0)={\eta}_{3}(0)={\tau}_{2}$ *satisfies the end conditions* ${\eta}_{1}({\theta}_{max})={\eta}_{2}({\theta}_{max})={\tau}_{1}$, ${\eta}_{3}({\theta}_{max})=\pi /2$.

*Proof*First we show that ${\eta}_{2}({\theta}_{max})={\tau}_{1}$ is satisfied. Integration of the last row of (21) from $\theta =0$ to $\theta ={\theta}_{max}$ using (26) gives

Because $\mathcal{I}(t)>0$ for all *t* except for two points at the boundary, we can conclude ${\eta}_{2}(\theta )={\tau}_{1}$. Note that this implies ${\tau}_{1}\le {\eta}_{2}(\theta )\le {\tau}_{2}$.

*t*except at the boundary points, we conclude

□

## 4 Numerical procedure and results

We solved the mathematical model described in the previous section to design three different TIR collimators. The collimators were designed for two different LEDs, which we refer to as LED16 and LED02. Both of them are Luxeon Rebel IES white LEDs without a dichroic coating, and have a larger than usual CoA variation. The intensity and chromaticity-coordinates of the LEDs were measured, and the measured data were interpolated. The interpolation polynomials have been used to approximate $\mathcal{I}(t)$ and $y(t)$ in (22a)-(22c). The first two collimators were designed for LED16 and have a Gaussian-shaped target intensity profile. The two collimators differ in their values for ${\tau}_{1}$ and ${\tau}_{2}$. The third collimator was designed for LED02 and has a block-shaped target intensity profile. The collimators were evaluated using the LightTools software package [13].

### 4.1 Modelling of the LEDs

*t*and 4 different angles

*u*. For each LED, the chromaticity values were averaged and the intensities were summed over the angle

*u*. These data have been interpolated with a least squares fit using the following polynomials:

**Coefficients from the linear least squares fits**

LED | i | 0 | 2 | 3 | 4 | 5 | 6 | 7 |
---|---|---|---|---|---|---|---|---|

16 | ${C}_{i}$ | 0 | −181.8279 | 76.0797 | 221.9411 | −624.0461 | 499.8253 | −127.0787 |

${D}_{i}^{x}$ | 0.4013 | 0.0578 | −0.0367 | 0.0271 | −0.0564 | 0.0444 | −0.0116 | |

${D}_{i}^{y}$ | 0.3546 | 0.0988 | −0.0557 | 0.0255 | −0.0371 | 0.0196 | −0.0034 | |

02 | ${C}_{i}$ | 0 | −148.8533 | −60.3797 | 520.5934 | −913.6760 | 636.8623 | −153.5863 |

${D}_{i}^{x}$ | 0.4591 | 0.0452 | −0.0590 | 0.0916 | −0.1399 | 0.1030 | −0.0271 | |

${D}_{i}^{y}$ | 0.3792 | 0.0892 | −0.1790 | 0.3596 | −0.4585 | 0.2823 | −0.0658 |

*t*was discretized into 46 different subintervals, labeled $k=1,2,\dots ,46$. For $k=1$ we have the interval $(0,\frac{\pi}{180})$, for $k=2,\dots ,45$ we have $(\frac{(2k-3)\pi}{180},\frac{(2k-1)\pi}{180})$ and finally for $k=46$ we have $(\frac{89\pi}{180},\frac{\pi}{2})$. The intensity and chromaticity coordinates of the LED models in these subintervals correspond to the measured data of the real LEDs at the angles $t=\frac{(2k-2)\pi}{180}$. The size of the LED model was reduced to 0.01 mm by 0.01 mm to simulate a point light source. A comparison of the measured data, the least squares fit and the raytracing results of the LightTools model of LED16 without collimator can be seen in Figures 5 and 6. A scatter plot of the measured

*x*and

*y*chromaticity coordinates for this LED was shown earlier in Figure 2. The plot shows the near-linear relationship between

*x*and

*y*, indeed.

### 4.2 Computation of the transfer functions

with $0\le \theta \le 1.25{\theta}_{\mathrm{FWHM}}={\theta}_{max}$, ${\theta}_{\mathrm{FWHM}}=\pi /9$. The collimator was designed for LED16. The choice of ${\tau}_{1}$ and ${\tau}_{2}$ is restricted by (29). This relation is highly nonlinear. A scatter plot of values of ${\tau}_{1}$ and ${\tau}_{2}$ that satisfy (29) for LED16 is shown in Figure 4.

**Parameter values and characteristics for the three different collimators**

Collimator | LED | ${\mathit{\tau}}_{\mathbf{1}}$ | ${\mathit{\tau}}_{\mathbf{2}}$ | r | ${\mathit{y}}_{\mathit{T}}$ | c | ${\mathit{\theta}}_{\mathbf{max}}$ |
---|---|---|---|---|---|---|---|

Gaussian, small 2nd segment | LED16 | 0.20 | 0.25 | 0.1630 | 0.3862 | 5,585.9 | 11.25 |

Gaussian, large 2nd segment | LED16 | 0.16 | 0.30 | 0.4417 | 0.3862 | 5,585.9 | 11.25 |

Block profile | LED02 | 0.20 | 0.25 | 0.1612 | 0.3986 | 4,843.3 | 9 |

### 4.3 Performance of the TIR collimators

**Average chromaticity coordinates of the LEDs and the maximum difference with the chromaticity coordinates in the simulations**

Collimator | LED | Target x | max(|Δx|) | Target y | max(|Δy|) |
---|---|---|---|---|---|

Gaussian, small 2nd segment | LED16 | 0.4181 | 4⋅10 | 0.3862 | 7⋅10 |

Gaussian, large 2nd segment | LED16 | 0.4181 | 4⋅10 | 0.3862 | 1⋅10 |

Block profile | LED02 | 0.4691 | 4⋅10 | 0.3985 | 5⋅10 |

## 5 Conclusions

We introduced an inverse method to design a TIR collimator that eliminates CoA variation for a point light source. This method improves the method introduced earlier in [6] by producing collimators that closely resemble standard collimators and at the same time have more parameters for optical design. In Section 3 we discussed which choices for these design parameters give meaningful results. In Section 4 we tested the method and verified the resulting collimators with Monte-Carlo raytracing using the software package LightTools. The simulations show color variations that are not visible with the human eye.

Unfortunately, LEDs are too large to be treated as a point light source. In future research, we would like to extend this method to take the finite size of the light source into account using iterative methods such as described in for example [18, 19]. This point source method will be an important building block in such an iterative method.

## Declarations

## Authors’ Affiliations

## References

- Haitz R, Jeffrey YT:
**Solid-state lighting: ‘The case’ 10 years after and future prospects.***Phys Status Solidi A*2011,**208:**17–29. 10.1002/pssa.201026349View ArticleGoogle Scholar - Schubert F:
*Light-Emitting Diodes*. 2nd edition. Cambridge University Press, Cambridge; 2006.View ArticleGoogle Scholar - Wu H, Narendran N, Gu Y, Bierman A:
**Improving the performance of mixed-color white LED systems by using scattered photon extraction technique. Proc. SPIE 6669.***Light-Emitting Diodes*2007.Google Scholar - Mueller GO: Luminescent ceramic for a light emitting device. United States patent US7361938 B2.Google Scholar
- Wang K, Wu D, Chen F, Liu Z, Luo X, Liu S:
**Angular color uniformity enhancement of white light-emitting diodes integrated with freeform lenses.***Opt Lett*2010,**35:**1860–1862. 10.1364/OL.35.001860View ArticleGoogle Scholar - Prins C, ten Thije Boonkkamp C, Tukker T, IJzerman W:
**An inverse method for the design of TIR collimators to achieve a uniform color light beam.***J Eng Math*2013,**81:**177–190. 10.1007/s10665-012-9584-7MathSciNetView ArticleGoogle Scholar - Maes M:
**Mathematical methods for reflector design**.*PhD thesis*. University of Amsterdam; 1997. Maes M: Mathematical methods for reflector design. PhD thesis. University of Amsterdam; 1997.Google Scholar - Malacara D:
*Color Vision and Colorimetry. Theory and Applications*. SPIE, Bellingham; 2002.Google Scholar - Wyszecki G, Stiles W:
*Color Science. Concepts and Methods, Quantitative Data and Formulae*. 2nd edition. Wiley, New York; 2000.Google Scholar - Bortz J, Shatz N:
**Generalized functional method of nonimaging optical design. Proc. SPIE 6338.**In*Nonimaging Optics and Efficient Illumination Systems III*Edited by: Winston R, Benitez P. 2006.Google Scholar - Bortz J, Shatz N:
**Mathematical relationships between the generalized functional, edge-ray and sms design methods. Nonimaging Optics: Efficient Design for Illumination and Solar Concentration 7785.***Nonimaging Optics and Efficient Illumination Systems III*2010.Google Scholar - Stöcker H:
*Taschenbuch mathematischer Formeln und moderner Verfahren*. Verlag Harri Deutsch, Thun; 1993.MATHGoogle Scholar **ORA LightTools product website**[http://www.opticalres.com/lt/ltprodds_f.html] [Accessed May 31, 2013].**Instrument Systems - Light measurement**[http://www.instrumentsystems.com/products/led-measurement-accessories/ledgon-goniophotometer/] [Accessed July 10, 2012].- Weisstein E:
**Gaussian function**[http://mathworld.wolfram.com/GaussianFunction.html] - Weisstein E:
**Full width at half maximum**[http://mathworld.wolfram.com/FullWidthatHalfMaximum.html] - MacAdam D:
**Visual sensitivities to color differences in daylight.***J Opt Soc Am*1942,**32:**247–274. 10.1364/JOSA.32.000247View ArticleGoogle Scholar - Zhang W, Liu Q, Gao H, Yu F: Free-form reflector optimization for general lighting. Opt Eng 2010.,49(6): Article ID 063003 Article ID 063003Google Scholar
- Cassarly WJ:
**Iterative reflector design using a cumulative flux compensation approach.***Opt Eng*2010.,**7652:**Article ID 76522 Article ID 76522Google Scholar

## Copyright

This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.