PULSE ANALYZER

Abstract

There is described a pulse analyser (1) for sampling a pulse or a repetitive stream of pulses. The pulse analyser multiplies a pulse by a set of basis functions (c1, c2) to generate a plurality of multiplied pulse functions, and a synthesiser (9a, 9b, 13) combines the multiplied pulse functions to generate a pulse sample. In particular, the synthesiser performs at least one integrating operation over an integration interval substantially corresponding to the duration of the pulse and at least one adding operation. The basis functions are such that the output of the synthesiser corresponds to a pulse sample for a sample time interval shorter than the integration interval.

Description

FIELD OF THE INVENTION

This invention relates to a method and apparatus for obtaining simultaneously discrete-time samples of both a wideband transient pulse and a quadrature signal associated with the pulse; the method being especially, but not exclusively, applicable to electronic systems, including oscilloscopes, capable of locating a pulse in time, determining two quadrature representations of the pulse, and extracting parameters characterizing the pulse shape.

BACKGROUND OF THE INVENTION

Suppose that a wideband pulse x(t) of finite duration and unknown shape is to be sampled at a plurality of J time instants t₁, t₂, . . . , t_j, . . . t_J. It is assumed that the pulse duration is limited by some maximum value T, and that the pulse time-of-arrival is approximately known. The acquired samples of the pulse x(t) are then used to determine some pulse descriptors such as shape and its moments, including location and time spread. The pulse under examination may be regarded as being observed at the output of a suitable sensor that has captured a portion of electromagnetic radiation scattered by a remote object of interest.

A review of the development of sampling techniques is given in “50 Years of RF and Microwave Sampling” by Mark Kahrs, IEEE Trans. Microwave Theory Tech., vol. 51, no. 1, pp. 1787-1804, June 2003.

Conventional sampling techniques utilise ultra-fast sampling circuits to produce instantaneous signal samples. However, such ultra-fast sampling circuits are generally expensive.

SUMMARY OF THE INVENTION

Aspects of the invention are set out in the accompanying claims.

The method according to the invention does not use ‘time slices’ of a pulse, but instead processes the whole pulse to obtain its ‘instantaneous’ values. Consequently, one of the distinct advantages of the invention follows from its potential capability of producing ‘instantaneous’ signal samples without actually using expensive ultra-fast sampling circuits. Each ‘instantaneous’ sample is obtained by suitably combining various averages determined over the duration T of the entire pulse x(t).

BRIEF DESCRIPTION OF THE DRAWINGS

Various embodiments of the invention will now be described with reference to the attached figures, in which:

FIG. 1 depicts an example of a pulse x(t) being sampled at time t_j with the use of a function approximating the Dirac impulse δ(t);

FIG. 2 shows the shape of a periodic sampling function D₈(t) obtained by combining a constant and eight consecutive harmonics;

FIG. 3 depicts schematically the operations to be performed to obtain a single sample x(t_j) of a pulse x(t);

FIG. 4 is an example of a pulse x(t) being sampled at time t_i with the use of a function approximating the kernel γ(t)=1/(πt);

FIG. 5 shows the shape of a quadrature sampling function H₈(t) obtained by combining eight consecutive harmonics;

FIG. 6 depicts a first new sampling function A₈(t) constructed in accordance with the invention;

FIG. 7 depicts a second new sampling function B₈(t) constructed in accordance with the invention; and

FIG. 8 is a functional block diagram of a pulse analyzer PAN constructed in accordance with the invention.

DESCRIPTION OF THE INVENTION

Suppose that a wideband pulse x(t) of finite duration and unknown shape is to be sampled at a plurality of J time instants t₁, t₂, . . . , t_j, . . . , t_J. It is assumed that the pulse duration is limited by some maximum value T, and that the pulse time-of-arrival is approximately known. The acquired samples of the pulse x(t) are then used to determine some pulse descriptors such as shape and its moments, including location and time spread. The pulse under examination may be regarded as being observed at the output of a suitable sensor that has captured a portion of electromagnetic radiation scattered by a remote object of interest.

From the ‘sifting’ property of the Dirac delta function, or Dirac impulse, δ(t) it follows that a sample at time t_j of a pulse x(t), i.e. the value x(t_j), can be determined from the integral

$x\left(t_j\right)={\int }_0^Tx\left(t\right)\delta \left(t-t_j\right)t$

FIG. 1 depicts an example of a pulse x(t) being sampled at time t_j with the use of a function δ′(t) approximating the Dirac impulse δ(t).

The Dirac delta function δ(t) can be approximated by a central segment of a sampling function of the form

$D_K\left(t\right)=a_0+\sum _{k=1}^Ka_k\cos 2\pi f_kt$

where

{a₀,a₁,a₂, . . . , a_k, . . . , a_K}

is a set of predetermined (K+1) coefficients, and

{f₁,f₂, . . . , f_k, . . . , f_K}

is a corresponding set of predetermined K frequencies.

It is convenient, although not necessary, to utilize frequencies f_k that are successive harmonics of some suitably chosen basic frequency f_B, i.e.

f_k=k f_B,k=1,2, . . . , K

For illustrative purposes, FIG. 2 shows the shape of a periodic sampling function D₈(t) obtained by combining a constant and eight consecutive harmonics, where

D ₈(t)=0.117+0.226 cos t+0.201 cos 2t+0.164 cos 3t+0.123 cos 4t++0.083 cos 5t+0.050 cos 6t+0.025 cos 7t+0.010 cos 8t

The above sampling function approximates the Dirac delta function δ(t) within the time interval (−π, π) equal to the period of the lowest used frequency, 1/(2π). In order to retain just one peak, the sampling function is multiplied by a unit-amplitude time gate g(t), spanning the time interval (−π, π).

In the above design, the peak width, i.e. the parameter FWHH (full width at half height) is equal to 90 percent of the period of the highest frequency used; the magnitude of the sidelobe level is kept below 0.0037 (i.e. −48.6 dB).

In general, the required peak width (FWHH) will depend on time (or range) resolution of the sensing system. For example, if the time resolution is equal to 1 ns (equivalent to range resolution of 0.15 m), then the peak width of D_K(t) should not exceed 1 ns; consequently, in a design similar to the illustrative example above, the highest frequency f_K should not be less than 900 MHz.

For correct operation, the period of the lowest used frequency f₁ should, at least, be equal to the pulse duration T. For example, if the expected pulse duration is T=30 ns, then the lowest frequency f₁ should not exceed 33 MHz.

In accordance with a first aspect of the invention, a sample at time t_j of pulse x(t), i.e. the value x(t_j), is determined by implementing Procedure 1:

1. selecting a first frequency f_k from a set of K predetermined frequencies

f₁,f₂, . . . , f_k, . . . , f_K

and generating a harmonic co-sinusoidal signal of the form

c _k(t_j)=cos(2πf_kt−θ_jk)

where θ_jk=2πf_kt_j is the initial phase;2. multiplying the pulse x(t) by a gated version of the harmonic signal c_k(t_j) and integrating the resulting product to obtain a value C_k(t_j);3. repeating steps 1 and 2 until each of the K frequencies has been utilized, such a repetitive operation resulting in a set of K values;

C₁(t_j),C₂(t_j), . . . , C_k(t_j), . . . , C_K(t_j)

4. integrating the pulse x(t) to obtain a constant value C₀ (this step is not needed, if the pulse x(t) has no DC (direct-current) component);5. utilizing such obtained values to form their linear combination as follows

a₀C₀+a₁C₁(t_j)+a₂C₂(t_j)+ . . . +a_kC_k(t_j)+ . . . +a_KC_K(t_j)

where {a₀, a₁, a₂, . . . , a_k, . . . , a_K} is a set of (K+1) predetermined coefficients; and6. using the above combination as an estimate of the sample x(t_j) of the pulse x(t) being analyzed.

FIG. 3 depicts schematically the operations to be performed in order to obtain a single sample x(t_j) of a pulse x(t) under examination.

Obviously, to obtain a suitable representation

{x(t₁),x(t₂), . . . , x(t_j), . . . , x(t_J)}

of a pulse x(t) being analyzed, Procedure 1 will have to be repeated for each of J time instants t_j, j=1, 2, . . . , J.

In contrast to conventional sampling, the disclosed method does not use ‘time slices’ of a pulse, but it processes the whole pulse to obtain its ‘instantaneous’ values. Consequently, one of the distinct advantages of the disclosed method follows from its potential capability of producing ‘instantaneous’ signal samples without actually using expensive ultra-fast sampling circuits. Each ‘instantaneous’ sample is obtained by suitably combining various averages determined over the duration T of the entire pulse x(t).

From the description of the disclosed method it follows that to obtain an accurate representation of a pulse x(t) under examination, the sampling function D_K(t) should be an adequate approximation of the Dirac delta function δ(t) over the entire interval T of pulse duration.

Various solutions to the above approximation problem are known to those skilled in the art, as problems of similar nature arise in spectral analysis and in the design of antenna arrays. A fairly comprehensive review of the available solutions can be obtained, for example, by studying the following publications:

1. F. J. Harris: “On the Use of Windows for Harmonic Analysis with the Discrete Fourier Transform,” Proc. IEEE, vol. 66, pp. 51-83, January 1978.2. A. H. Nuttall: “Some Windows with Very Good Sidelobe Behaviour,” IEEE Trans. Acoust. Speech Signal Processing, vol. ASSP-29, no. 1, pp. 84-91, February 1981.

3. C. A. Balanis, Antenna Theory. New York: Wiley, 1997.

4. N. Jin and Y. Rahmat-Samii: “Advances in Particle Swarm Optimization for Antenna Design,” IEEE Trans. Antennas Propag., vol. 55, no. 3, pp. 556-567, March 2007.

As known to those skilled in the art, a complete characterization of a wideband transient pulse x(t) can only be obtained by determining additionally the so-called quadrature signal y(t) associated with the underlying pulse x(t). In principle, a quadrature signal can be produced by a Hilbert transformer that convolves an input pulse with kernel γ(t)=1/(πt); such an operation is usually performed digitally by suitable processing of discrete-time samples of the input pulse.

If the quadrature signal y(t) is available, then the pulse x(t) of interest can be characterized by its (Hilbert) envelope z(t) and phase function φ(t), defined by

z(t)√{square root over (x²(t)+y²(t))}{square root over (x²(t)+y²(t))}

φ(t)tan⁻¹[y(t)/x(t)]

where tan⁻¹(·) is a four-quadrant function. The envelope z(t) may then be used to determine the pulse position in time by estimating some location parameters, such as the mean value ('centre of gravity'), median or the ‘dominating’ mode.

For example, the mean location ('centre of gravity') of a pulse is calculated from

t=∫tz ²(t)dt/∫z²(t)dt

It should be noted that the above formula uses the squared envelope z²(t), sometimes referred to as ‘power’ distribution.

The median location t_M of a pulse is defined as

${\int }_0^{t_M}z^2\left(t\right)t=\frac{1}{2}{\int }_0^{\infty }z^2\left(t\right)t$

Hence, the median location t_M is a time instant so selected within the pulse duration as to obtain equal energy in the left and the right portions of the pulse x(t).

The mode location t_P of a pulse is defined as the time instant at which the power z²(t) of the pulse x(t) reaches its maximum value, hence

$t_p=\underset{t}{\max }\left\{z^2\left(t\right)\right\}$

From the above discussion it follows that the knowledge of the wideband pulse x(t) itself together with supplementary information provided by its associated quadrature signal y(t) allows determination of the pulse location.

When the pulse location in time has been determined, it may be advantageous to assign the total energy ∫z²(t) dt of that pulse to this particular location. Such an operation may be viewed as pulse condensation or pulse compaction.

In some applications, it may be of interest to determine higher moments of the pulse power distribution z²(t) by using the methods known to those skilled in the art. In particular, the second central moment will characterize the pulse spread in time, whereas the third central moment will provide information regarding the pulse ‘skewness’.

When x(t) is a pulse scattered by a complex object of interest, the envelope z(t) itself, or the power distribution in time z²(t), will supply some information about the object's structure. In some cases, also the phase function φ(t) will be used to provide complementary information.

In a way similar to sampling with the Dirac impulse, a sample at time t_i of the quadrature signal y(t) associated with the underlying pulse x(t), i.e. the value y(t_i), can be determined from the integral

$y\left(t_i\right)={\int }_0^Tx\left(t\right)\gamma \left(t-t_i\right)t$

where γ(t)=1/(πt) is a Hilbert transform of the Dirac impulse δ(t). (Because of the singularity of γ(t), the principal value of the above integral must be used.)

FIG. 4 is an example of a pulse x(t) being sampled at time t_i with the use of a function H_K(t) approximating the kernel γ(t)=1/(πt). It should be noted that such sampling of the pulse x(t) produces in fact samples of the associated quadrature signal y(t).

In accordance with another aspect of the invention, kernel γ(t)=1/(πt) is approximated by a central segment of a quadrature sampling function of the form

$H_K\left(t\right)=\sum _{k=1}^Ka_k\sin 2\pi f_kt$

where both the sets of predetermined K coefficients and frequencies, {a₁, a₂, a_k, . . . , a_K} and {f₁, f₂, . . . , f_k, . . . , f_K}, have the same elements as the corresponding sets used for constructing a sampling function D_K(t).

For illustrative purposes, FIG. 5 shows the shape of a quadrature sampling function H₈(t) obtained by combining eight consecutive harmonics, where

H ₈(t)=0.226 sin t+0.201 sin 2t+0.164 sin 3t+0.123 sin 4t++0.083 sin 5t+0.050 sin 6t+0.025 sin 7t+0.010 sin 8t

It should be noted that the above quadrature sampling function H₈(t) is a Hilbert transform of the sampling function D₈(t), depicted in FIG. 2.

The quadrature sampling function approximates the kernel γ(t) within the time interval (−π, π) equal to the period of the lowest used frequency, 1/(2π). In order to select the correct segment of the function, the sampling function is multiplied by a unit-amplitude time gate g(t), spanning the time interval (−π, π).

In accordance with a further aspect of the invention, a sample at time t_i of the quadrature signal y(t) associated with pulse x(t), i.e. the value y(t_i), is determined by implementing Procedure 2:

1. selecting a first frequency f_k from a set of K predetermined frequencies

f₁,f₂, . . . , f_k, . . . , f_K

and generating a harmonic sinusoidal signal of the form

s _k(t_i)=sin(2πf_kt−θ_ik)

where θ_ik=2πf_kt_i is the initial phase;2. multiplying the pulse x(t) by a gated version of the harmonic signal s_k(t_i) and integrating the resulting product to obtain a value S_k(t_i);3. repeating steps 1 and 2 until each of the K frequencies has been utilized; such a repetitive operation will result in a set of K values

S₁(t_i),S₂(t_i), . . . , S_k(t_i), . . . , S_K(t_i)

4. utilizing such obtained values to form their linear combination as follows

a₁S₁(t_i)+a₂S₂(t_i)+ . . . +a_kS_k(t_i)+ . . . +a_KS_K(t_i)

where {a₁, a₂, . . . , a_k, . . . a_K} is a set of K predetermined coefficients; and5. using the above combination as an estimate of the sample y(t_i) of the quadrature signal y(t) associated with the pulse x(t) under examination.

Obviously, in order to obtain a suitable representation

{y(t₁),y(t₂), . . . , y(t_i), . . . , y(t₁)}

of the quadrature signal y(t) associated with pulse x(t), Procedure 2 will have to be repeated for each of I time instants t_i, where i=1, 2, . . . , I.

A further distinct advantage of the disclosed aspects of the invention follows from its potential ability to produce ‘instantaneous’ samples of the quadrature signal y(t) by sampling in fact an underlying pulse x(t). Those samples are determined without the use of expensive ultra-fast sampling circuits and complicated digital signal processing. Each ‘instantaneous’ sample of the quadrature signal is obtained by suitably combining various averages determined over the entire duration T of the underlying pulse x(t).

Although Procedure 1 and Procedure 2 have similar structures, the ranges of the corresponding time indices, {1, 2, . . . , j, . . . , J} and {1, 2, . . . , i, . . . , I}, are very different. This discrepancy follows from different shapes of the sampling function D_K(t) and the quadrature sampling function H_K(t). The sampling function D_K(t) approximates an impulse and, therefore, is concentrated within a short time interval, whereas the time extent of the quadrature sampling function H_K(t) is intentionally large (compare FIG. 22 and FIG. 25). Those dramatically different time scales of the sampling functions D_K(t) and H_K(t) make concurrent running of Procedures 1 and 2 a very difficult task indeed. However, it would be advantageous to perform sampling of an underlying pulse and an associated quadrature signal at the same time instants. Therefore, for this purpose some other two sampling functions need to be devised.

In accordance with a further aspect of the invention, the sampling functions D_K(t) and H_K(t) are used to construct two new sampling functions, A_K(t) and B_K(t), defined as follows

A _K(t)=[D_K(t)+H_K(t)]√{square root over (2)}

B _K(t)=[D_K(t)−H_K(t)]√{square root over (2)}

FIG. 6 and FIG. 7 depict two new sampling functions, A₈(t) and B₈(t), obtained from previously considered functions D₈(t) and H₈(t) and multiplied by a suitable unit-amplitude time gate. As seen, the two shapes are mirror images of one another and are, therefore, of the same time scale.

The disclosed construction results in the following forms of the two new sampling functions A_K(t) and B_K(t)

$A_K\left(t\right)=\frac{a_0}{\sqrt{2}}+\sum _{k=1}^Ka_k\cos \left(2\pi f_kt-\pi /4\right)$ $\mathrm{and}$ $B_K\left(t\right)=\frac{a_0}{\sqrt{2}}+\sum _{k=1}^Ka_k\cos \left(2\pi f_kt+\pi /4\right)$

The new sampling functions, A_K(t) and B_K(t), may be viewed as a result of rotation by π/4 of the original sampling functions D_K(t) and H_K(t).

In accordance with yet another aspect of the invention, the two new sampling functions, A_K(t) and B_K(t), are used to sample an underlying pulse x(t) to obtain samples of its two images (representations), u(t) and v(t). Those images, being in quadrature to one another, will preserve the shape of the envelope z(t) of the pulse x(t), i.e.,

z(t)=√{square root over (x²(t)+y²(t))}{square root over (x²(t)+y²(t))}=√{square root over (u²(t)+v²(t))}{square root over (u²(t)+v²(t))}

Consequently, all descriptors of the original pulse x(t), including shape and mean location, can be determined in the usual manner.

Furthermore, the new phase function ψ(t)

ψ(t)=tan⁻¹[v(t)/u(t)]

will be a replica of the original phase function φ(t) shifted by a constant π/4. Sample at time t_j of two representations, u(t) and v(t), of the underlying pulse x(t), i.e. the values u(t_j) and u(t_j), are determined by implementing Procedure 3:1. selecting a first frequency f_k from a set of K predetermined frequencies

f₁,f₂, . . . , f_k, . . . , f_K

and generating two harmonic co-sinusoidal signal of the form

c ₁ k(t_j)=cos(2πf_kt−θ_jk−π/4

c ₂ k(t_j)=cos(2πf_kt−θ_jk+π/4)

where θ_jk=2πf_kt_j is the initial phase;2. multiplying the pulse x(t) by gated versions of the two harmonic signals, c_1k(t_j) and c_2k(t_j), and integrating the resulting products to obtain two values: U_k(t_j) and V_k(t_j);3. repeating steps 1 and 2 until each of the K frequencies has been utilized; such a repetitive operation will result in a set of (2K) values

U₁(t_j),U₂(t_j), . . . , U_k(t_j), . . . , U_K(t_j)

V₁(t_j),V₂(t_j), . . . , V_k(t_j), . . . , V_K(t_j)

4. integrating the pulse x(t) to obtain a constant value C₀ (this step is not needed, if the pulse x(t) has no DC (direct-current) component);5. forming two linear combinations as follows

a₀C₀/√2+a₁U₁(t_j)+a₂U₂(t_j)+ . . . +a_kU_k(t_j)+ . . . +a_KU_K(t_j)

a₀C₀/√2+a₁V₁(t_j)+a₂V₂(t_j)+ . . . +a_kV_k(t_j)+ . . . +a_KV_K(t_j)

where {a₀, a₁, a₂, . . . , a_k, . . . , a_K} is a set of (K+1) predetermined coefficients; and6. using the above combinations as respective estimates of the samples u(t_j) and v(t_j) of the two representations u(t) and v(t) of the pulse x(t) being analyzed.

Obviously, in order to obtain suitable representations

{u(t₁),u(t₂), . . . , u(t_j), . . . , u(t_J)}

{v(t₁),v(t₂), . . . , v(t_j), . . . , v(t_J)}

of the two ‘images’, u(t) and v(t), of the underlying pulse x(t), Procedure 3 will have to be repeated for each of J time instants t_j, where j=1, 2, . . . , J.

FIG. 8 is a functional block diagram of a pulse analyzer (PAN) 1 constructed in accordance with the invention. The analyzer PAN implements Procedure 3 disclosed above.

The system comprises a signal conditioning circuit (SCC) 3, an optical-fibre recirculating loop (RCL) 5, a first (MXC) 7a, a second mixer (MXS) 7b, a first integrator (AVC) 9a, a second integrator (AVS) 9b, a direct digital synthesizer (DDS) 11, an arithmetic unit (ARM) 13, and a timing/control unit (TCU) 15.

The signal conditioning circuit 3 captures a single pulse x(t) that appears transiently at input XX and sends the pulse to the recirculating loop 5 that regenerates this pulse to produce, at output XR, a pulse train comprising a plurality of replicas of the pulse x(t). The recirculating loop 5 also produces a synchronizing signal SN, preceding each of the pulse replicas.

Each of the identical pulses constituting the pulse train is applied in parallel to the two mixers (multipliers) 7a and 7b. The other mixer inputs are driven by respective co-sinusoidal signals, C1 and C2, supplied by the direct digital synthesizer 11: mixer 7a receives signal C1 of the form

c ₁ k(t_j)=cos(2πf_kt−θ_jk−π/4)

whereas mixer 7b is driven by signal C2

c ₂ k(t_j)=cos(2πf_kt−θ_jk+π/4)

In the above formulae, f_k is a frequency selected from a set of K predetermined frequencies, f₁, f₂, . . . , f_k, . . . , f_K, and the initial phase is determined from

θ_jk=2πf_kt_j

where t_j is a time instant at which a sample is taken.

The output signals supplied by the mixers 7a and 7b, are respectively applied to the two gated integrators 9a and 9b via their respective inputs, PC and PS. The integrators 9 perform integration of their input signals, PC and PS, during a time interval determined by a time gate GT supplied by the timing/control unit 15. The values, AC and AS, produced by the integrators 9 are then sent to the arithmetic unit 13.

The arithmetic unit 13 utilizes input values, AC and AS, produced for each of K predetermined frequencies, f₁, f₂, . . . , f_k, . . . , f_K, and for each of J predetermined time instants, t₁, t₂, . . . , t_j, . . . , t_J, to determine the pulse shape, its envelope z(t) and phase function ψ(t) and, if required, other parameters of interest, such as mean time location, time spread etc. For correct identification of the actual frequency f_k and time instant t_j, the arithmetic unit 13 receives, from the timing/control unit 15 via input FT, a frequency index f and the time index j.

The direct digital synthesizer 11 produces two signals, C1 and C2, required by the mixers 7a and 7b in response to two control signals, FR and PH, used by the synthesizer 11 to set the correct frequency, f_k, and phase θ_jk.

In general, the recirculating loop 5 has to produce (J·K) identical replicas of the input transient pulse x(t). For example, if the number K of used frequencies is 32 and the number J of time instants is also 32, the loop 5 has to supply 1032 replicas. This is a realistic requirement; for example, a system described in: Yan Yin, Beam Diagnostics with Optical-Fibre Optics. Proc. 2005 Particle Accelerator Conf., Knoxville, pp. 3040-30-42, May 2005, is capable of producing 3000 replicas with a 2-km long optical-fibre loop. See also:

• • Ming-Chiang Li, RF Signal Train Generator and Interferoceivers.
  • U.S. Pat. No. RE37,561, 26 Feb. 2002.
  • Yan Yin, Method and Device for Measuring Single-Shot Transient Signals.
  • U.S. Pat. No. 6,738,133, 18 May 2004.

Yan Yin and Xiuge Che, New Developments on Single-Shot Fibre Scope. Proc. 2006 European Particle Accelerator Conf., Edinburgh, pp. 1253-1255, 2006.

Instead of using a recirculating loop, a single pulse could be split and input into multiple parallel channels for processing.

The foregoing description of preferred embodiments of the invention has been presented for the purpose of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. In light of the foregoing description, it is evident that many alterations, modifications, and variations will enable those skilled in the art to utilize the invention in various embodiments suited to the particular use contemplated.

Claims

1. A pulse analyser for sampling a pulse or a repetitive stream of pulses, the pulse analyser comprising: multiplying means operable to multiply a pulse by a set of basis functions to generate a plurality of multiplied pulse functions; anda synthesiser operable to combine the multiplied pulse functions to generate a pulse sample,wherein the synthesiser is operable to perform at least one integrating operation over an integration interval substantially corresponding to the duration of the pulse and at least one adding operation, and wherein the basis functions are such that the output of the synthesiser corresponds to a pulse sample for a sample time interval shorter than the integration interval.

2. A pulse analyser according to claim 1, wherein the basis functions are selected so that the pulse analyser, in effect, multiplies the pulse or repetitive stream of pulses by an approximation of a delta function at a timing in the centre of the sample time interval.

3. A pulse analyser according to claim 2, wherein the delta function is an analytic delta function.

4. A pulse analyser according to claim 2, wherein the delta function has components corresponding to an approximation of a Dirac delta function and an approximation of a Hilbert transform of the Dirac delta function.

5. A pulse analyser according to claim 2, wherein the delta function has two components, each component corresponding to a respective different combination of an approximation of a Dirac delta function and an approximation of a Hilbert transform of the Dirac delta function.

6. A pulse analyser according to claim 5, wherein the two components form a mirrored pair of sampling functions.

7. A pulse analyser according to claim 1, wherein the basis functions are selected so that the pulse analyser, in effect, multiplies the pulse or repetitive stream of pulses by an approximation of the Hilbert transform of a delta function centred in the sample time interval.

8. A pulse analyser according to claim 1, wherein the synthesiser is operable to integrate each of the multiplied pulse functions over the integration interval to generate a plurality of integration values, and to perform an adding operation on the integration values.

9. A pulse analyser according to claim 8, wherein said adding operation comprises a weighted addition.

10. A pulse analyser according to claim 1, further comprising a signal conditioner operable to capture a single pulse and to generate a repetitive stream of pulses in which each pulse is a replica of the captured pulse.

11. A pulse analyser according to claim 10, wherein the signal conditioner comprises a recirculating loop having an input for receiving a captured pulse and an output for outputting said repetitive stream of pulses.

12. A pulse analyser according to claim 11, wherein said input is operable to convert an electrical pulse into a corresponding optical pulse, said recirculating loop comprises a loop of optical fiber and an amplifier, and the output comprise means for converting part of the optical signal circulating in the optical fibre loop into a corresponding electrical signal.

13. A pulse analyser according to claim 1, wherein the basis functions are oscillating signals at respective different frequencies.

14. A pulse analyser according to claim 1, wherein the pulse analyser is operable to sample the pulse or repetitive stream of pulses at a plurality of different timings to generate an envelope function for the pulse.

15. A pulse analyser according to claim 14, further comprising means for determining a property of the pulse from the pulse envelope.

16. A pulse analyser according to claim 15, further comprising means for determining a timing for the pulse based on said envelope function.

17. A pulse analyser according to claim 16, wherein the determining means calculates at least one of the mean timing, the median timing and the mode timing for the pulse based on said envelope function.

18. A pulse analyser for generating a sample of a pulse by multiplying the pulse with a plurality of functions and integrating over an integration interval, wherein said plurality of functions form a sampling function for a time interval shorter than the integration interval.

19. A method of analysing a pulse, the method comprising applying an analysis function to the pulse by deriving a sample value for each of plural time instants within the duration of the pulse, each sample value being derived by combining the pulse with each of a plurality of spectral components of a version of the analysis function for the respective time instant, and deriving a combination of the results integrated over at least a substantial portion of the pulse duration.

20. A method as claimed in claim 19, comprising applying first and second analysis functions to the pulse, the first and second functions being orthogonal or having orthogonal components.