1: \documentclass[preprint,showpacs,preprintnumbers,amsmath,amssymb,nofootinbib]{revtex4}

2: 

3: %\documentclass[twocolumn, showpacs]{revtex4}

4: 

5: \usepackage{graphicx}

6: \usepackage{dcolumn}% Align table columns on decimal point

7: \usepackage{bm}% bold math

8: 

9: \begin{document}

10: \title{\bf On the laser frequency stabilization by locking to a LISA arm}

11: 

12: \author{Massimo Tinto}

13: \email{Massimo.Tinto@jpl.nasa.gov}

14: \altaffiliation [Also at: ]{Space Radiation Laboratory, California

15:   Institute of Technology, Pasadena, CA 91125}

16: \affiliation{Jet Propulsion Laboratory, California Institute of

17: Technology, Pasadena, CA 91109} 

18: 

19: \author{Malik Rakhmanov}

20: \email{malik@phys.ufl.edu}

21: \affiliation{Department of Physics, University of Florida,

22:   Gainesville, FL 32611}

23: 

24: \date{\today} 

25: \begin{abstract}

26:   LISA is an array of three spacecraft flying in an approximately

27:   equilateral triangle configuration, which will be used as a

28:   low-frequency detector of gravitational waves. Recently a technique

29:   has been proposed for suppressing the phase noise of the onboard

30:   lasers by locking them to the LISA arms. In this paper we show that

31:   the delay-induced effects substantially modify the performance of

32:   this technique, making it different from the conventional locking 

33:   of lasers to optical resonators. We analyze these delay-induced 

34:   effects in both transient and steady-state regimes and discuss their 

35:   implications for the implementation of this technique on LISA.

36: \end{abstract}

37: 

38: \pacs{04.80.Nn, 95.55.Ym, 07.60.Ly}

39: \maketitle

40: 

41: 

42: 

43: 

44: The Laser Interferometer Space Antenna (LISA) \cite{1} is a

45: space-borne interferometer that will use coherent laser beams

46: exchanged between three widely separated spacecraft to detect

47: low-frequency ($10^{-4}-1$ Hz) gravitational waves. Each spacecraft is

48: carrying two optical benches with lasers, optics, photo-detectors, and

49: drag-free proof masses. It has been shown that the time series of

50: Doppler shifts of the laser beams between spacecraft pairs, and those

51: between adjacent optical benches within each spacecraft, can be

52: combined, with suitable time delays, to cancel the otherwise

53: overwhelming laser phase noise.  This post-processing data technique

54: is known as time-delay interferometry (TDI) (see \cite{2} and

55: references therein).  Recently a method to reduce the laser phase

56: noise at the time of detection by using the arms of LISA has been

57: proposed by Sheard et al.  \cite{3}.  In this one-arm locking

58: technique a fraction of the beam from the laser on board of one

59: spacecraft interferes with a beam coherently retransmitted back by

60: another spacecraft, as shown in Fig. 1.  The phase difference of the

61: two beams forms an error signal for a control system feeding back to

62: the laser.  Originally proposed as an alternative to the TDI

63: technique, this method can be implemented in conjunction with TDI to

64: provide laser pre-stabilization, which will relax the requirements for

65: implementing the TDI technique.

66: 

67: 

68: The motivation for the one-arm stabilization method comes from the

69: conventional techniques of locking lasers to optical resonators, in

70: which the suppression of laser frequency noise is ultimately defined

71: by the stability of the resonator. In the case of LISA, this notion

72: becomes substantially modified by the delay-induced effects

73: originating from the time of flight within the LISA arm. Most

74: noteworthy of these effects is the prolonged ringdown discovered in 

75: numerical simulations \cite{3} and later confirmed in table-top

76: experiments \cite{Garcia04}. In this paper we provide a general theory 

77: for these delay-induced effects and analyze their impact on the 

78: performance of this stabilization technique.

79: 

80: 

81: % In this paper we provide a theory for these delay-induced 

82: % effects and discuss their implications for operation of LISA.

83: 

84: % In this paper we study these delay-induced effects 

85: % for two cases of closed loop dynamics: steady-state and transient regimes.

86: 

87: % In this paper we show that these delay-induced effects are not

88: % artifacts of a particular implementation of the control system but

89: % are fundamental, and discuss their implications for operation of LISA.

90: 

91: 

92: 

93: Following \cite{3}, we neglect variations in the distance between the

94: spacecraft and thus assume that the duration of the photon round-trip

95: within the LISA arm ($\tau$) is constant: $\tau = 33.3$~s.  The

96: inverse of the photon round-trip within the LISA arm ($1/\tau)$ plays

97: a special role in this analysis and will be called free spectral range

98: (FSR) by analogy with optical resonators \cite{Verdeyen}. Our notations

99: are shown in Fig. 1: $p(t)$ is the free-running laser phase noise,

100: $q(t)$ is the residual phase noise, $r(t)$ is the error signal, and

101: $G(t)$ is the filtering function of the control system. The

102: time-domain equations for the closed-loop dynamics are

103: %

104: \begin{eqnarray}

105: %

106:    r(t) & = & q(t) - q(t - \tau),

107:               \label{def:r(t)} \\

108: %

109:    q(t) & = & p(t) - \int_0^t G(t-t') \; r(t') \, dt' ,

110:               \label{def:q(t)}

111: %

112: \end{eqnarray}

113: %

114: where we explicitly show that the feedback loop is closed at time

115: $t=0$. 

116: 

117: \begin{figure}[t]

118:    \centering\includegraphics[width=0.8\textwidth]{Fig1.eps}

119:    \caption{Schematic diagram of the one-arm laser stabilization 

120:      technique (adapted from \cite{3}).}

121:    \label{diagram}

122: \end{figure}

123: 

124: Consistency requires the laser to be running at least for the duration

125: of one photon round-time before the loop is closed, implying:

126: \begin{equation}

127: q(t) = p(t) \ \ \ , \ \ \ {\rm  for} \ t < 0 \ .

128: \label{boundary}

129: \end{equation}

130: By taking the Laplace transform of Eq.(\ref{def:r(t)}),

131: (\ref{def:q(t)}), and properly accounting for equation

132: (\ref{boundary}), we finally get

133: %

134: \begin{eqnarray}

135: %

136:    r(s) & = & q(s) - e^{-s \tau} \; q(s) - 

137:           \int_0^{\tau} e^{-st} \; p(t - \tau) \; dt ,\\

138: %

139:    q(s) & = & p(s) - G(s) \; r(s) .

140: %

141: \end{eqnarray}

142: %

143: Eliminating $r(s)$, we find that the residual noise in the

144: laser frequency stabilization loop is given by

145: %

146: \begin{equation}

147: %

148:    q(s) = C(s) \; p(s) + 

149:       C(s) \; G(s) \; \int_0^{\tau} e^{-st} \; p(t - \tau) \; dt ,

150: %

151: \label{closedLoop}

152: \end{equation}

153: %

154: where $C(s)$ is the closed-loop transfer function:

155: %

156: \begin{equation}

157: %

158:    C(s) = \frac{1}{1 + G(s)(1 - e^{-s \tau})} .

159: %

160: \end{equation}

161: %

162: Notice that the last term in Eq.(\ref{closedLoop}) is missing in the

163: analysis of \cite{3}. We now consider transient and steady-state

164: dynamics which are described by Eq.(\ref{closedLoop}). Henceforth we

165: assume that the feedback loop is stable and therefore all the

166: poles of $C(s)$ are on the left-half-plane of the Laplace domain.

167: 

168: Mismatch of the initial conditions (non-zero value of the error

169: signal) at the moment of closing the feedback loop causes start-up

170: transients \cite{3, Garcia04}, which can hinder the performance of the

171: one-arm locking. Here we provide simple formulas for the frequencies 

172: $f_n$ and relaxation times $\tau_n$ of these transients. They can

173: be found from the real and imaginary parts of the poles of the

174: closed-loop transfer function:

175: %

176: \begin{equation}

177: %

178:    s_n \equiv 2 \pi i f_n  - \frac{1}{\tau_n} .

179: %

180: \end{equation}

181: %

182: The closed loop transfer function $C(s)$ has poles

183: which are defined by the characteristic equation:

184: %

185: \begin{equation}

186:   1 - e^{-s \tau} = - \frac{1}{G(s)}.

187: \label{Charac}

188: \end{equation}

189: %

190: This equation has an infinite number of roots typical of dynamical 

191: systems with delay \cite{5}. For large gain, the roots can be found 

192: perturbatively in terms of inverse powers of $G$. Setting the right

193: hand side of Eq.(\ref{Charac}) to zero we obtain the zero-order 

194: approximation:

195: %

196: \begin{equation}

197: %

198:    \bar{s}_n = \frac{2\pi i}{\tau} \; n  ,

199: %

200: \end{equation}

201: %

202: where $n$ is integer. The next order approximation would be $s_n =

203: \bar{s}_n + \delta s_n$, where $\delta s_n$ is a small correction

204: vanishing in the limit of infinite gain. Simple algebra shows that

205: $\delta s_n \approx - [\tau G(\bar{s}_n)]^{-1}$.  Therefore, to first

206: order in $G^{-1}$, the poles of the closed-loop transfer function are

207: given by

208: %

209: \begin{equation}

210: %

211:    s_n = \frac{2 \pi i}{\tau} \; n - \frac{1}{\tau G(\bar{s}_n)}.

212: %

213: \end{equation}

214: %

215: Then the frequencies and relaxation times of the transients are

216: %

217: \begin{eqnarray}

218: %

219:    f_n & = & \frac{n}{\tau} - \frac{{\mathrm{Im}}\{

220:              G(\bar{s}_n)^{-1}\}}{2\pi\tau} ,\label{freqn} \\

221: %

222:    \tau_n & = & \frac{\tau}{{\mathrm{Re}} \{ 

223:                 G(\bar{s}_n)^{-1} \} } . \label{taun}

224: \end{eqnarray}

225: %

226: Thus, we have shown that the transients form damped oscillations with 

227: frequencies fairly close to multiples of the free spectral range, and

228: their relaxation times are defined by the open-loop gain of the

229: control system. The amplitude of these transients can vary from lock

230: to lock, depending on the values of $p(0)$ and $p(-\tau)$. They can be 

231: significantly reduced by introducing a ramping function to the control

232: system which slowly increases the gain from zero to the designed value 

233: \cite{3,Garcia04}. 

234: However, the frequencies $f_n$ and relaxation times $\tau_n$ of the

235: transients are defined by the properties of the system and not the

236: initial conditions. To avoid prolonged ringdown of the transients, 

237: one can reduce $\tau_n$ by properly designing 

238: the open loop gain $G(s)$.

239: 

240: 

241: 

242: We now turn our attention to the steady-state dynamics of the closed

243: loop. Using the identity,

244: %

245: \begin{equation}

246: %

247:    C(s) \; G(s) = \frac{1 - C(s)}{1 - e^{-s \tau}} ,

248: %

249: \end{equation}

250: %

251: we can rewrite Eq.(\ref{closedLoop}) in the following equivalent

252: form:

253: %

254: \begin{equation}

255: %

256:    q(s) = C(s) \left[p(s) - P(s) \right] + P(s),

257:    \label{suppNoise}

258: %

259: \end{equation}

260: %

261: where we introduced a new quantity:

262: %

263: \begin{equation}

264:    P(s) = \frac{\int\limits_0^{\tau} e^{-st} \; 

265:       p(t - \tau) \; dt}{1 - e^{- s \tau}} .

266: \label{periodic}

267: \end{equation}

268: %

269: This is the Laplace transform of a periodic function of time \cite{4}, 

270: $P(t)$, which is determined by the repetition of the first $\tau$ seconds of the

271: free-running laser noise $p(t)$

272: %

273: \begin{equation}

274: %

275:    P(t + n \tau) = p(t) ,

276: %

277: \end{equation}

278: %

279: where $-\tau < t \leq 0$ and $n$ is integer.

280: % (For Laplace transforms of periodic functions of time see for example

281: % \cite{4}.)

282: Equation (\ref{suppNoise}) means that the free-running laser phase noise is 

283: compared with the periodic replica of its first $\tau$ seconds and the

284: difference between the two is suppressed by the loop gain $G(s)$. For

285: very large gain, $C(s)$ becomes negligible and

286: Eq.(\ref{suppNoise}) reduces to 

287: %

288: \begin{equation}

289: %

290:    q(s) = P(s) \ ,

291:    \label{limit}

292: %

293: \end{equation}

294: %

295: which describes the limiting performance of the one-arm locking

296: technique. In this case, the phase noise during the first $\tau$ 

297: seconds repeats itself indefinitely, thus becoming a periodic function

298: of time. The same conclusion can be drawn already from 

299: Eq.(\ref{def:r(t)}) which shows that the null point of the control 

300: system: $r(t) = 0$ implies periodicity of the residual noise: 

301: $q(t)=q(t - \tau)$. 

302: 

303: 

304: The above calculation is based on the assumption of constant

305: round-trip-light-time which corresponds to perfect length reference. In this

306: situation, conventional stabilization of lasers to optical 

307: resonators would lead to an unlimited suppression of the laser 

308: phase noise. (In practice, the limitations would come from other 

309: noise sources such as the shot noise in the photo-detectors, which we

310: neglected in this analysis.)

311: % In practice, the limitations would come from other 

312: % noise sources such as the shot noise in the photo-detectors or 

313: % any other noise produced by imperfections of sensor or actuator

314: % designs. For the sake of simplicity we neglect such sources of

315: % noise in this discussion. 

316: As we have shown, the laser phase noise in the one-arm locking method

317: approaches a limit which is independent from the stability of the LISA

318: arm. Since the free-running laser phase noise is compared against

319: itself (at some earlier time) it can only be suppressed to the level

320: of its own stability within the first $\tau$ seconds.  This still

321: implies a suppression when viewed in the frequency domain.  Namely,

322: the power spectrum of the laser phase noise is no longer spread over a

323: large range of frequencies, rather it is concentrated within the

324: harmonics of the FSR. For finite gains, the peaks appear at

325: frequencies $f_n$ defined by the poles $s_n$ and are slightly offset

326: from the exact harmonics of the FSR. Furthermore, the peaks will have

327: a finite width and quality factors $Q_n = \pi f_n \tau_n$. Periodicity

328: of the residual noise suggests filtering methods which can done either

329: during detection (notch filtering) or in post-processing.

330: \vskip12pt\noindent

331: M.R. acknowledges support from the US National Science Foundation under 

332: grant PHY-0244902. This research was performed at the Jet Propulsion 

333: Laboratory, California Institute of Technology, under contract with 

334: the National Aeronautics and Space Administration.

335: 

336: \begin{references}

337: \bibitem{1} P.L. Bender, K.  Danzmann, and the LISA Study Team,

338: {\it{Laser \ Interferometer \ Space \ Antenna \ for \ the \ Detection \ of

339: \ Gravitational \ Waves, \ Pre-Phase \ A \ Report}},

340: {\bf{MPQ 233}} (Max-Planck-Instit\"ut f\"ur

341: Quantenoptik, Garching), July 1998.

342: \bibitem{2} M. Tinto, F.B. Estabrook, and J.W. Armstrong, {\it Phys.

343:     Rev. D}, {\bf 69}, 082001 (2004)

344: \bibitem{3} B.S. Sheard, M.B. Gray, D.E. McClelland, and

345:     D.A. Shaddock, {\it Phys. Lett. A}, {\bf 320}, 9 (2003).

346: \bibitem{Garcia04} A.F. Garcia Marin, G. Heinzel, R. Schilling,

347:   V. Wand, F.G. Cervantes, F. Steier, O. Jenrich, A. Weidner, and

348:   K. Danzmann, {\it Laser Phase Locking to a LISA Arm: Experimental

349:     Approach}. To appear in: {\it Proceedings of the 5th LISA

350:     Symposium, ESTEC, Noordwijk, The Netherlands}, to be published in 

351:    {\it Class. Quantum Grav.} (2004).

352: \bibitem{Verdeyen} J.T. Verdeyen, {\it Laser Electronics}, 3rd

353:   edition, Prentice Hall, New Jersey (1994).

354: \bibitem{4} R.J. Beerends, {\it et al.}, {\it Fourier and Laplace

355:     Transforms}, Cambridge University Press, Cambridge (UK), (2003) 

356: \bibitem{5} L.E. El'sgol'ts, {\it Introduction to the Theory of

357:     Differential Equations with Deviating Arguments}, Holden-Day,

358:     Inc., San-Francisco, (1966) 

359: \end{references}

360: \end{document}

361: 

362: